It wasn’t always so clear, but the Rust programming language is fundamentally about empowerment: no matter what kind of code you are writing now, Rust empowers you to reach farther, to program with confidence in a wider variety of domains than you did before.

Take, for example, “systems-level” work that deals with low-level details of memory management, data representation, and concurrency. Traditionally, this realm of programming is seen as arcane, accessible only to a select few who have devoted the necessary years learning to avoid its infamous pitfalls. And even those who practice it do so with caution, lest their code be open to exploits, crashes, or corruption.

Rust breaks down these barriers by eliminating the old pitfalls and providing a friendly, polished set of tools to help you along the way. Programmers who need to “dip down” into lower-level control can do so with Rust, without taking on the customary risk of crashes or security holes, and without having to learn the fine points of a fickle toolchain. Better yet, the language is designed to guide you naturally towards reliable code that is efficient in terms of speed and memory usage.

Programmers who are already working with low-level code can use Rust to raise their ambitions. For example, introducing parallelism in Rust is a relatively low-risk operation: the compiler will catch the classical mistakes for you. And you can tackle more aggressive optimizations in your code with the confidence that you won’t accidentally introduce crashes or exploits.

But Rust isn’t limited to low-level systems programming. It’s expressive and ergonomic enough to make CLI apps, web servers, and many other kinds of code quite pleasant to write — you’ll find simple examples of both later in the book. Working with Rust allows you to build skills that transfer from one domain to another; you can learn Rust by writing a web app, then apply those same skills to target your Raspberry Pi.

This book fully embraces the potential of Rust to empower its users. It’s a friendly and approachable text intended to help you level up not just your knowledge of Rust, but also your reach and confidence as a programmer in general. So dive in, get ready to learn—and welcome to the Rust community!

— Nicholas Matsakis and Aaron Turon

Welcome to “The Rust Programming Language,” an introductory book about Rust.

Rust is a programming language that helps you write faster, more reliable software. High-level ergonomics and low-level control are often at odds with each other in programming language design; Rust stands to challenge that. Through balancing powerful technical capacity and a great developer experience, Rust gives you the option to control low-level details (such as memory usage) without all the hassle traditionally associated with such control.

Rust is great for many people for a variety of reasons. Let’s discuss a few of the most important groups.

Rust is proving to be a productive tool for collaborating among large teams of developers with varying levels of systems programming knowledge. Low-level code is prone to a variety of subtle bugs, which in most other languages can only be caught through extensive testing and careful code review by experienced developers. In Rust, the compiler plays a gatekeeper role by refusing to compile code with these kinds of bugs--including concurrency bugs. By working alongside the compiler, the team can spend more time focusing on the logic of the program rather than chasing down bugs.

Rust also brings contemporary developer tools to the systems programming world:

• Cargo, the included dependency manager and build tool, makes adding, compiling, and managing dependencies painless and consistent across the Rust ecosystem.
• Rustfmt ensures a consistent coding style across developers.
• The Rust Language Server powers IDE integration for code completion and inline error messages.

By using these and other tools in the Rust ecosystem, developers can be productive while writing systems-level code.

Rust is for students and people who are interested in learning about systems concepts. Many people have learned about topics like operating systems development through Rust. The community is happy to answer student questions. Through efforts such as this book, the Rust teams want to make systems concepts more accessible to more people, especially those getting started with programming.

Rust is used in production by hundreds of companies, large and small, for a variety of tasks, such as command line tools, web services, DevOps tooling, embedded devices, audio and video analysis and transcoding, cryptocurrencies, bioinformatics, search engines, internet of things applications, machine learning, and even major parts of the Firefox web browser.

Rust is for people who want to build the Rust programming language, community, developer tools, and libraries. We’d love for you to contribute to the Rust language.

By speed, we mean both the speed of the programs that Rust lets you create and the speed at which Rust lets you write them. The Rust compiler’s checks ensure stability through feature additions and refactoring, as opposed to brittle legacy code in languages without these checks that developers are afraid to modify. By striving for zero-cost abstractions, higher level features that compile to lower level code as fast as code written manually, Rust endeavors to make safe code be fast code as well.

This isn’t a complete list of everyone the Rust language hopes to support, but these are some of the biggest stakeholders. Overall, Rust’s greatest ambition is to take trade-offs that have been accepted by programmers for decades and eliminate the dichotomy. Safety and productivity. Speed and ergonomics. Give Rust a try, and see if its choices work for you.

This book assumes that you’ve written code in some other programming language, but doesn’t make any assumptions about which one. We’ve tried to make the material broadly accessible to those from a wide variety of programming backgrounds. We don’t spend a lot of time talking about what programming is or how to think about it; someone new to programming entirely would be better served by reading a book specifically providing an introduction to programming.

This book generally assumes that you’re reading it front-to-back, that is, later chapters build on top of concepts in earlier chapters, and earlier chapters may not dig into details on a topic, revisiting the topic in a later chapter.

There are two kinds of chapters in this book: concept chapters, and project chapters. In concept chapters, you’ll learn about an aspect of Rust. In the project chapters, we’ll build small programs together, applying what we’ve learned so far. Chapters 2, 12, and 20 are project chapters; the rest are concept chapters.

Additionally, Chapter 2 is a hands-on introduction to Rust as a language. We’ll cover concepts at a high level, and later chapters will go into them in detail. If you’re the kind of person who likes to get their hands dirty right away, Chapter 2 is great for that. If you’re really that kind of person, you may even wish to skip over Chapter 3, which covers features that are very similar to other programming languages, and go straight to Chapter 4 to learn about Rust’s ownership system. By contrast, if you’re a particularly meticulous learner who prefers to learn every detail before moving onto the next, you may want to skip Chapter 2 and go straight to Chapter 3.

Chapter 5 discusses structs and methods, and Chapter 6 covers enums, match expressions, and the if let control flow construct. Structs and enums are the ways to make custom types in Rust.

In Chapter 7, you'll learn about Rust's module system and privacy for organizing your code and its public API. Chapter 8 discusses some common collection data structures provided by the standard library: vectors, strings, and hash maps. Chapter 9 is all about Rust's error handling philosophy and techniques.

Chapter 10 digs into generics, traits, and lifetimes, which give you the power to define code that applies to multiple types. Chapter 11 is all about testing, which is still necessary even with Rust's safety guarantees to ensure your program's logic is correct. In Chapter 12, we'll build a subset of the functionality of the grep command line tool that searches for text within files and we'll use many of the concepts we discussed in the previous chapters.

Chapter 13 explores closures and iterators: features of Rust that come from functional programming languages. In Chapter 14, we'll explore more about Cargo and talk about best practices for sharing your libraries with others. Chapter 15 discusses smart pointers provided by the standard library and the traits that enable their functionality.

In Chapter 16, we'll go through different models of concurrent programming and how Rust helps you to program using multiple threads fearlessly. Chapter 17 looks at how Rust idioms compare to Object Oriented Programming principles you may be familiar with.

Chapter 18 is a reference on patterns and pattern matching, which are powerful ways of expressing ideas throughout Rust programs. Chapter 19 is a smorgasbord of advanced topics that you might be interested in, including unsafe Rust and more about lifetimes, traits, types, functions, and closures.

In Chapter 20, we'll finish up with a project where we'll implement a low-level multithreaded web server!

Finally, there are some appendices. These contain useful information about the language in a more reference-like format.

In the end, there’s no wrong way to read a book: if you want to skip ahead, go for it! You may have to jump back if you find things confusing. Do whatever works for you.

An important part of the process of learning Rust is learning how to read the error messages that the compiler gives you. As such, we’ll be showing a lot of code that doesn’t compile, and the error message the compiler will show you in that situation. As such, if you pick a random example, it may not compile! Please read the surrounding text to make sure that you didn’t happen to pick one of the in-progress examples.

This book is open source. If you find an error, please don’t hesitate to file an issue or send a pull request on GitHub. Please see CONTRIBUTING.md for more details.

Let’s get your Rust journey started! In this chapter, we’ll discuss:

• Installing Rust on Linux, Mac, or Windows
• Writing a program that prints “Hello, world!”
• Using cargo, Rust’s package manager and build system

The first step to using Rust is to install it. We’ll download Rust through rustup, a command-line tool for managing Rust versions and associated tools. For this you’ll need an internet connection.

The following steps will install the latest stable version of the Rust compiler. The examples and output shown in this book all use stable Rust 1.21.0. Rust’s stability guarantees ensure that all of the examples in the book that compile will continue to compile with newer versions of Rust. The output may differ slightly between versions, as error messages and warnings are often improved. In other words, any newer, stable version of Rust you will install with these steps should work as expected with the content of this book.

Command Line Notation

In this chapter and throughout the book we’ll be showing some commands used in the terminal. Lines that should be entered in a terminal all start with $. You don’t need to type in the $ character, it is simply there to indicate the start of each command. Many tutorials use this convention: $ for commands run as a regular user, and # for commands you should be running as an administrator. Lines that don’t start with $ are typically showing the output of the previous command. Additionally, PowerShell specific examples will use > rather than $.

If you’re on Linux or a Mac, open a terminal and enter the following command:

$ curl https://sh.rustup.rs -sSf | sh

This will download a script and start the installation of the rustup tool, which installs the latest stable version of Rust. You may be prompted for your password. If it all goes well, you’ll see this appear:

Rust is installed now. Great!

Of course, if you distrust using curl URL | sh to install software, you can download, inspect, and run the script however you like.

The installation script automatically adds Rust to your system PATH after your next login. If you want to start using Rust right away instead of restarting your terminal, run the following command in your shell to add Rust to your system PATH manually:

$ source $HOME/.cargo/env

Alternatively, you can add the following line to your ~/.bash_profile:

$ export PATH="$HOME/.cargo/bin:$PATH"

Finally, you’ll need a linker of some kind. It’s likely you already have one installed, but if you try to compile a Rust program and get errors telling you that a linker could not be executed, you’ll need to install one. You can install a C compiler, as that will usually come with the correct linker. Check your platform’s documentation for how to install a C compiler. Some common Rust packages depend on C code and will need a C compiler too, so it may be worth installing one now regardless.

On Windows, go to https://www.rust-lang.org/en-US/install.html and follow the instructions for installing Rust. At some point in the installation you’ll receive a message telling you you’ll also need the C++ build tools for Visual Studio 2013 or later. The easiest way to acquire the build tools is to install Build Tools for Visual Studio 2017, found in the Other Tools and Frameworks section.

The rest of this book will use commands that work in both cmd.exe and PowerShell. If there are specific differences, we’ll explain which to use.

If you have reasons for preferring not to use rustup, please see the Rust installation page for other options.

Once you have Rust installed via rustup, updating to the latest version is easy. From your shell, run the update script:

$ rustup update

To uninstall Rust and rustup, from your shell, run the uninstall script:

$ rustup self uninstall

To check whether you have Rust installed correctly, open up a shell and enter:

$ rustc --version

You should see the version number, commit hash, and commit date for the latest stable version at the time you install in the following format:

rustc x.y.z (abcabcabc yyyy-mm-dd)

If you see this, Rust has been installed successfully! Congrats!

If you don’t and you’re on Windows, check that Rust is in your %PATH% system variable.

If that’s all correct and Rust still isn’t working, there are a number of places you can get help. The easiest is the #rust IRC channel on irc.mozilla.org, which you can access through Mibbit. Go to that address, and you’ll be chatting with other Rustaceans (a silly nickname we call ourselves) who can help you out. Other great resources include the Users forum and Stack Overflow.

The installer also includes a copy of the documentation locally, so you can read it offline. Run rustup doc to open the local documentation in your browser.

Any time there’s a type or function provided by the standard library and you’re not sure what it does or how to use it, use the API (Application Programming Interface) documentation to find out!

Now that you have Rust installed, let’s write your first Rust program. It’s traditional when learning a new language to write a little program to print the text “Hello, world!” to the screen, so we’ll do the same here!

Note: This book assumes basic familiarity with the command line. Rust itself makes no specific demands about your editing, tooling, or where your code lives, so if you prefer an IDE (Integrated Development Environment) to the command line, feel free to use your favorite IDE. Many IDEs now have some degree of Rust support; check the IDE’s documentation for details. Enabling great IDE support has been a recent focus of the Rust team, and progress has been made rapidly on that front!

First, make a directory to put your Rust code in. Rust doesn’t care where your code lives, but for the exercises and projects in this book, we’d suggest making a projects directory in your home directory and keeping all your projects there.

Open a terminal and enter the following commands to make a projects directory and, inside that, a directory for this “Hello, world!” project:

Linux and Mac:

$ mkdir ~/projects
$ cd ~/projects
$ mkdir hello_world
$ cd hello_world

Windows CMD:

> mkdir "%USERPROFILE%\projects"
> cd /d "%USERPROFILE%\projects"
> mkdir hello_world
> cd hello_world

Windows PowerShell:

> mkdir $env:USERPROFILE\projects
> cd $env:USERPROFILE\projects
> mkdir hello_world
> cd hello_world

Next, make a new source file and call it main.rs---Rust files always end with the .rs extension. If you’re using more than one word in your filename, use an underscore to separate them. For example, you’d use hello_world.rs rather than helloworld.rs.

Now open the main.rs file you just created, and enter the code shown in Listing 1-1:

Filename: main.rs

fn main() {
    println!("Hello, world!");
}

Listing 1-1: A program that prints “Hello, world!”

Save the file, and go back to your terminal window. On Linux or macOS, enter the following commands to compile and run the file:

$ rustc main.rs
$ ./main
Hello, world!

On Windows, use .\main.exe instead of ./main.

> rustc main.rs
> .\main.exe
Hello, world!

Regardless of your operating system, you should see the string Hello, world! print to the terminal. If you don’t see this output, see the “Troubleshooting” section earlier for ways to get help.

If you did see Hello, world! printed, then congratulations! You’ve officially written a Rust program. That makes you a Rust programmer! Welcome!

Now, let’s go over what just happened in your “Hello, world!” program in detail. Here’s the first piece of the puzzle:

fn main() {

}

These lines define a function in Rust. The main function is special: it is always the first code that is run for every executable Rust program. The first line declares a function named main that has no parameters and returns nothing. If there were parameters, their names would go inside the parentheses, ( and ).

Also note that the function body is wrapped in curly brackets, { and }. Rust requires these around all function bodies. It’s considered good style to place the opening curly bracket on the same line as the function declaration, with one space in between.

At the time of writing, an automatic formatter, rustfmt, is under development. If you’d like to stick to a standard style across Rust projects, rustfmt is a tool that will format your code in a particular style. The plan is to eventually include it with the standard Rust distribution, like rustc, so depending on when you read this book, you may have it already installed! Check the online documentation for more details.

Inside the main function, we have this code:

# #![allow(unused_variables)]
#fn main() {
    println!("Hello, world!");
#}

This line does all of the work in this little program: it prints text to the screen. There are a number of details to notice here. The first is that Rust style is to indent with four spaces, not a tab.

The second important detail is the println! call. This code is calling a Rust macro. If it were calling a function instead, it would be entered as println (without the !). We’ll discuss Rust macros in more detail in Appendix D, but for now you just need to know that when you see a ! that means that you’re calling a macro instead of a normal function.

Next comes"Hello, world!" which is a string. We pass this string as an argument to println! and the total effect is that the string is printed to the screen. Easy enough!

We end the line with a semicolon ;, which indicates that this expression is over, and the next one is ready to begin. Most lines of Rust code end with a ;.

You’ve just seen how to run a newly created program, so now let’s break that process down and examine each step.

Before running a Rust program, you have to compile it using the Rust compiler by entering the rustc command and passing it the name of your source file, like this:

$ rustc main.rs

If you come from a C or C++ background, you’ll notice that this is similar to gcc or clang. After compiling successfully, Rust outputs a binary executable.

On Linux, Mac, and PowerShell on Windows, you can see the executable by entering the ls command in your shell as follows:

$ ls
main  main.rs

With CMD on Windows, you’d enter:

> dir /B %= the /B option says to only show the file names =%
main.exe
main.pdb
main.rs

This shows we have two files: the source code, with the .rs extension, and the executable (main.exe on Windows, main everywhere else). All that’s left to do from here is run the main or main.exe file, like this:

$ ./main  # or .\main.exe on Windows

If main.rs were your “Hello, world!” program, this would print Hello, world! to your terminal.

If you come from a dynamic language like Ruby, Python, or JavaScript, you may not be used to compiling and running a program being separate steps. Rust is an ahead-of-time compiled language, which means that you can compile a program, give the executable to someone else, and they can run it even without having Rust installed. If you give someone a .rb, .py, or .js file, on the other hand, they need to have a Ruby, Python, or JavaScript implementation installed (respectively), but you only need one command to both compile and run your program. Everything is a tradeoff in language design.

Just compiling with rustc is fine for simple programs, but as your project grows, you’ll want to be able to manage all of the options and make it easy to share your code. Next, we’ll introduce you to a tool called Cargo, which will help you write real-world Rust programs.

Cargo is Rust’s build system and package manager. Most Rustaceans will use this tool to manage their Rust projects because Cargo takes care of a lot of tasks for you, such as building your code, downloading the libraries your code depends on, and building those libraries. (We call libraries your code needs dependencies.)

The simplest Rust programs, like the one we’ve written so far, don’t have any dependencies, so if we had built the Hello World project with Cargo, it would only be using the part of Cargo that takes care of building your code. As you write more complex Rust programs, you’ll want to add dependencies, and if you start the project off using Cargo, that will be a lot easier to do.

As the vast majority of Rust projects use Cargo, the rest of this book will assume that you’re using Cargo too. Cargo comes installed with Rust itself, if you used the official installers as covered in the “Installation” section. If you installed Rust through some other means, you can check if you have Cargo installed by entering the following into your terminal:

$ cargo --version

If you see a version number, great! If you see an error like command not found, then you should look at the documentation for your method of installation to determine how to install Cargo separately.

Let’s create a new project using Cargo and look at how it differs from our original Hello World project. Navigate back to your projects directory (or wherever you decided to put your code) and then on any operating system run:

$ cargo new hello_cargo --bin
$ cd hello_cargo

This creates a new binary executable called hello_cargo. The --bin argument to passed to cargo new makes an executable application (often just called a binary), as opposed to a library. We’ve given hello_cargo as the name for our project, and Cargo creates its files in a directory of the same name.

Go into the hello_cargo directory and list the files, and you should see that Cargo has generated two files and one directory for us: a Cargo.toml and a src directory with a main.rs file inside. It has also initialized a new git repository, along with a .gitignore file.

Note: Git is a common version control system. You can change cargo new to use a different version control system, or no version control system, by using the --vcs flag. Run cargo new --help to see the available options.

Open up Cargo.toml in your text editor of choice. It should look similar to the code in Listing 1-2:

Filename: Cargo.toml

[package]
name = "hello_cargo"
version = "0.1.0"
authors = ["Your Name <you@example.com>"]

[dependencies]

Listing 1-2: Contents of Cargo.toml generated by cargo new

This file is in the TOML (Tom’s Obvious, Minimal Language) format, which is what Cargo uses as its configuration format.

The first line, [package], is a section heading that indicates that the following statements are configuring a package. As we add more information to this file, we’ll add other sections.

The next three lines set the configuration information Cargo needs in order to know that it should compile your program: the name, the version, and who wrote it. Cargo gets your name and email information from your environment, so if that’s not correct, go ahead and fix that and save the file.

The last line, [dependencies], is the start of a section for you to list any of your project’s dependencies. In Rust, packages of code are referred to as crates. We won’t need any other crates for this project, but we will in the first project in Chapter 2, so we’ll use this dependencies section then.

Now open up src/main.rs and take a look:

Filename: src/main.rs

fn main() {
    println!("Hello, world!");
}

Cargo has generated a “Hello World!” for you, just like the one we wrote in Listing 1-1! So far, the differences between our previous project and the project generated by Cargo are that with Cargo our code goes in the src directory, and we have a Cargo.toml configuration file in the top directory.

Cargo expects your source files to live inside the src directory so that the top-level project directory is just for READMEs, license information, configuration files, and anything else not related to your code. In this way, using Cargo helps you keep your projects nice and tidy. There’s a place for everything, and everything is in its place.

If you started a project that doesn’t use Cargo, as we did with our project in the hello_world directory, you can convert it to a project that does use Cargo by moving the project code into the src directory and creating an appropriate Cargo.toml.

Now let’s look at what’s different about building and running your Hello World program through Cargo! From your project directory, build your project by entering the following commands:

$ cargo build
   Compiling hello_cargo v0.1.0 (file:///projects/hello_cargo)
    Finished dev [unoptimized + debuginfo] target(s) in 2.85 secs

This creates an executable file in target/debug/hello_cargo (or target\debug\hello_cargo.exe on Windows), which you can run with this command:

$ ./target/debug/hello_cargo # or .\target\debug\hello_cargo.exe on Windows
Hello, world!

Bam! If all goes well, Hello, world! should print to the terminal once more. Running cargo build for the first time also causes Cargo to create a new file at the top level called Cargo.lock, which is used to keep track of the exact versions of dependencies in your project. This project doesn’t have dependencies, so the file is a bit sparse. You won’t ever need to touch this file yourself; Cargo will manage its contents for you.

We just built a project with cargo build and ran it with ./target/debug/hello_cargo, but we can also use cargo run to compile and then run all in one go:

$ cargo run
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/hello_cargo`
Hello, world!

Notice that this time, we didn’t see the output telling us that Cargo was compiling hello_cargo. Cargo figured out that the files haven’t changed, so it just ran the binary. If you had modified your source code, Cargo would have rebuilt the project before running it, and you would have seen output like this:

$ cargo run
   Compiling hello_cargo v0.1.0 (file:///projects/hello_cargo)
    Finished dev [unoptimized + debuginfo] target(s) in 0.33 secs
     Running `target/debug/hello_cargo`
Hello, world!

Finally, there’s cargo check. This command will quickly check your code to make sure that it compiles, but not bother producing an executable:

$ cargo check
   Compiling hello_cargo v0.1.0 (file:///projects/hello_cargo)
    Finished dev [unoptimized + debuginfo] target(s) in 0.32 secs

Why would you not want an executable? cargo check is often much faster than cargo build, because it skips the entire step of producing the executable. If you’re checking your work throughout the process of writing the code, using cargo check will speed things up! As such, many Rustaceans run cargo check periodically as they write their program to make sure that it compiles, and then run cargo build once they’re ready to give it a spin themselves.

So to recap, using Cargo:

• We can build a project using cargo build or cargo check
• We can build and run the project in one step with cargo run
• Instead of the result of the build being put in the same directory as our code, Cargo will put it in the target/debug directory.

A final advantage of using Cargo is that the commands are the same no matter what operating system you’re on, so at this point we will no longer be providing specific instructions for Linux and Mac versus Windows.

When your project is finally ready for release, you can use cargo build --release to compile your project with optimizations. This will create an executable in target/release instead of target/debug. These optimizations make your Rust code run faster, but turning them on makes your program take longer to compile. This is why there are two different profiles: one for development when you want to be able to rebuild quickly and often, and one for building the final program you’ll give to a user that won’t be rebuilt repeatedly and that will run as fast as possible. If you’re benchmarking the running time of your code, be sure to run cargo build --release and benchmark with the executable in target/release.

With simple projects, Cargo doesn’t provide a whole lot of value over just using rustc, but it will prove its worth as you continue. With complex projects composed of multiple crates, it’s much easier to let Cargo coordinate the build.

Even though the hello_cargo project is simple, it now uses much of the real tooling you’ll use for the rest of your Rust career. In fact, to work on any existing projects you can use the following commands to check out the code using Git, change into the project directory, and build:

$ git clone someurl.com/someproject
$ cd someproject
$ cargo build

If you want to look at Cargo in more detail, check out its documentation.

You’re already off to a great start on your Rust journey! In this chapter, you’ve:

• Installed the latest stable version of Rust
• Written a “Hello, world!” program using both rustc directly and using the conventions of cargo

This is a great time to build a more substantial program, to get used to reading and writing Rust code. In the next chapter, we’ll build a guessing game program. If you’d rather start by learning about how common programming concepts work in Rust, see Chapter 3.

Let’s jump into Rust by working through a hands-on project together! This chapter introduces you to a few common Rust concepts by showing you how to use them in a real program. You’ll learn about let, match, methods, associated functions, using external crates, and more! The following chapters will explore these ideas in more detail. In this chapter, you’ll practice the fundamentals.

We’ll implement a classic beginner programming problem: a guessing game. Here’s how it works: the program will generate a random integer between 1 and 100. It will then prompt the player to enter a guess. After a guess is entered, the program will indicate whether the guess is too low or too high. If the guess is correct, the game will print a congratulatory message and exit.

To set up a new project, go to the projects directory that you created in Chapter 1 and make a new project using Cargo, like so:

$ cargo new guessing_game --bin
$ cd guessing_game

The first command, cargo new, takes the name of the project (guessing_game) as the first argument. The --bin flag tells Cargo to make a binary project, like the one in Chapter 1. The second command changes to the new project’s directory.

Look at the generated Cargo.toml file:

Filename: Cargo.toml

[package]
name = "guessing_game"
version = "0.1.0"
authors = ["Your Name <you@example.com>"]

[dependencies]

If the author information that Cargo obtained from your environment is not correct, fix that in the file and save it again.

As you saw in Chapter 1, cargo new generates a “Hello, world!” program for you. Check out the src/main.rs file:

Filename: src/main.rs

fn main() {
    println!("Hello, world!");
}

Now let’s compile this “Hello, world!” program and run it in the same step using the cargo run command:

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 1.50 secs
     Running `target/debug/guessing_game`
Hello, world!

The run command comes in handy when you need to rapidly iterate on a project, as we’ll do in this game, quickly testing each iteration before moving on to the next one.

Reopen the src/main.rs file. You’ll be writing all the code in this file.

The first part of the guessing game program will ask for user input, process that input, and check that the input is in the expected form. To start, we’ll allow the player to input a guess. Enter the code in Listing 2-1 into src/main.rs.

Filename: src/main.rs

use std::io;

fn main() {
    println!("Guess the number!");

    println!("Please input your guess.");

    let mut guess = String::new();

    io::stdin().read_line(&mut guess)
        .expect("Failed to read line");

    println!("You guessed: {}", guess);
}

Listing 2-1: Code that gets a guess from the user and prints it

This code contains a lot of information, so let’s go over it line by line. To obtain user input and then print the result as output, we need to bring the io (input/output) library into scope. The io library comes from the standard library (which is known as std):

use std::io;

By default, Rust brings only a few types into the scope of every program in the prelude. If a type you want to use isn’t in the prelude, you have to bring that type into scope explicitly with a use statement. Using the std::io library provides you with a number of useful features, including the ability to accept user input.

As you saw in Chapter 1, the main function is the entry point into the program:

fn main() {

The fn syntax declares a new function, the parentheses, (), indicate there are no parameters, and the curly bracket, {, starts the body of the function.

As you also learned in Chapter 1, println! is a macro that prints a string to the screen:

println!("Guess the number!");

println!("Please input your guess.");

This code is printing a prompt stating what the game is and requesting input from the user.

Next, we’ll create a place to store the user input, like this:

let mut guess = String::new();

Now the program is getting interesting! There’s a lot going on in this little line. Notice that this is a let statement, which is used to create a variable. Here’s another example:

let foo = bar;

This line creates a new variable named foo and bind it to the value bar. In Rust, variables are immutable by default. We’ll be discussing this concept in detail in the “Variables and Mutability” section in Chapter 3. The following example shows how to use mut before the variable name to make a variable mutable:

let foo = 5; // immutable
let mut bar = 5; // mutable

Note: The // syntax starts a comment that continues until the end of the line. Rust ignores everything in comments, which are discussed in more detail in Chapter 3.

Now you know that let mut guess will introduce a mutable variable named guess. On the other side of the equal sign (=) is the value that guess is bound to, which is the result of calling String::new, a function that returns a new instance of a String. String is a string type provided by the standard library that is a growable, UTF-8 encoded bit of text.

The :: syntax in the ::new line indicates that new is an associated function of the String type. An associated function is implemented on a type, in this case String, rather than on a particular instance of a String. Some languages call this a static method.

This new function creates a new, empty string. You’ll find a new function on many types, because it’s a common name for a function that makes a new value of some kind.

To summarize, the let mut guess = String::new(); line has created a mutable variable that is currently bound to a new, empty instance of a String. Whew!

Recall that we included the input/output functionality from the standard library with use std::io; on the first line of the program. Now we’ll call an associated function, stdin, on io:

io::stdin().read_line(&mut guess)
    .expect("Failed to read line");

If we hadn’t listed the use std::io line at the beginning of the program, we could have written this function call as std::io::stdin. The stdin function returns an instance of std::io::Stdin, which is a type that represents a handle to the standard input for your terminal.

The next part of the code, .read_line(&mut guess), calls the read_line method on the standard input handle to get input from the user. We’re also passing one argument to read_line: &mut guess.

The job of read_line is to take whatever the user types into standard input and place that into a string, so it takes that string as an argument. The string argument needs to be mutable so the method can change the string’s content by adding the user input.

The & indicates that this argument is a reference, which gives you a way to let multiple parts of your code access one piece of data without needing to copy that data into memory multiple times. References are a complex feature, and one of Rust’s major advantages is how safe and easy it is to use references. You don’t need to know a lot of those details to finish this program. For now, all you need to know is that like variables, references are immutable by default. Hence, you need to write &mut guess rather than &guess to make it mutable. (Chapter 4 will explain references more thoroughly.)

We’re not quite done with this line of code. Although what we’ve discussed so far is a single line of text, it’s only the first part of the single logical line of code. The second part is this method:

.expect("Failed to read line");

When you call a method with the .foo() syntax, it’s often wise to introduce a newline and other whitespace to help break up long lines. We could have written this code as:

io::stdin().read_line(&mut guess).expect("Failed to read line");

However, one long line is difficult to read, so it’s best to divide it: two lines for two method calls. Now let’s discuss what this line does.

As mentioned earlier, read_line puts what the user types into the string we’re passing it, but it also returns a value—in this case, an io::Result. Rust has a number of types named Result in its standard library: a generic Result as well as specific versions for submodules, such as io::Result.

The Result types are enumerations, often referred to as enums. An enumeration is a type that can have a fixed set of values, and those values are called the enum’s variants. Chapter 6 will cover enums in more detail.

For Result, the variants are Ok or Err. The Ok variant indicates the operation was successful, and inside Ok is the successfully generated value. The Err variant means the operation failed, and Err contains information about how or why the operation failed.

The purpose of these Result types is to encode error-handling information. Values of the Result type, like any type, have methods defined on them. An instance of io::Result has an expect method that you can call. If this instance of io::Result is an Err value, expect will cause the program to crash and display the message that you passed as an argument to expect. If the read_line method returns an Err, it would likely be the result of an error coming from the underlying operating system. If this instance of io::Result is an Ok value, expect will take the return value that Ok is holding and return just that value to you so you can use it. In this case, that value is the number of bytes in what the user entered into standard input.

If you don’t call expect, the program will compile, but you’ll get a warning:

$ cargo build
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
warning: unused `std::result::Result` which must be used
  --> src/main.rs:10:5
   |
10 |     io::stdin().read_line(&mut guess);
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   |
   = note: #[warn(unused_must_use)] on by default

Rust warns that you haven’t used the Result value returned from read_line, indicating that the program hasn’t handled a possible error.

The right way to suppress the warning is to actually write error handling, but since you just want to crash this program when a problem occurs, you can use expect. You’ll learn about recovering from errors in Chapter 9.

Aside from the closing curly brackets, there’s only one more line to discuss in the code added so far, which is the following:

println!("You guessed: {}", guess);

This line prints the string we saved the user’s input in. The set of curly brackets, {}, is a placeholder: think of {} as little crab pincers that hold a value in place. You can print more than one value using curly brackets: the first set of curly brackets holds the first value listed after the format string, the second set holds the second value, and so on. Printing multiple values in one call to println! would look like this:

# #![allow(unused_variables)]
#fn main() {
let x = 5;
let y = 10;

println!("x = {} and y = {}", x, y);
#}

This code would print x = 5 and y = 10.

Let’s test the first part of the guessing game. Run it using cargo run:

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 2.53 secs
     Running `target/debug/guessing_game`
Guess the number!
Please input your guess.
6
You guessed: 6

At this point, the first part of the game is done: we’re getting input from the keyboard and then printing it.

Next, we need to generate a secret number that the user will try to guess. The secret number should be different every time so the game is fun to play more than once. Let’s use a random number between 1 and 100 so the game isn’t too difficult. Rust doesn’t yet include random number functionality in its standard library. However, the Rust team does provide a rand crate.

Remember that a crate is a package of Rust code. The project we’ve been building is a binary crate, which is an executable. The rand crate is a library crate, which contains code intended to be used in other programs.

Cargo’s use of external crates is where it really shines. Before we can write code that uses rand, we need to modify the Cargo.toml file to include the rand crate as a dependency. Open that file now and add the following line to the bottom beneath the [dependencies] section header that Cargo created for you:

Filename: Cargo.toml

[dependencies]

rand = "0.3.14"

In the Cargo.toml file, everything that follows a header is part of a section that continues until another section starts. The [dependencies] section is where you tell Cargo which external crates your project depends on and which versions of those crates you require. In this case, we’ll specify the rand crate with the semantic version specifier 0.3.14. Cargo understands Semantic Versioning (sometimes called SemVer), which is a standard for writing version numbers. The number 0.3.14 is actually shorthand for ^0.3.14, which means “any version that has a public API compatible with version 0.3.14.”

Now, without changing any of the code, let’s build the project, as shown in Listing 2-2:

$ cargo build
    Updating registry `https://github.com/rust-lang/crates.io-index`
 Downloading rand v0.3.14
 Downloading libc v0.2.14
   Compiling libc v0.2.14
   Compiling rand v0.3.14
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 2.53 secs

Listing 2-2: The output from running cargo build after adding the rand crate as a dependency

You may see different version numbers (but they will all be compatible with the code, thanks to SemVer!), and the lines may be in a different order.

Now that we have an external dependency, Cargo fetches the latest versions of everything from the registry, which is a copy of data from Crates.io. Crates.io is where people in the Rust ecosystem post their open source Rust projects for others to use.

After updating the registry, Cargo checks the [dependencies] section and downloads any crates you don’t have yet. In this case, although we only listed rand as a dependency, Cargo also grabbed a copy of libc, because rand depends on libc to work. After downloading the crates, Rust compiles them and then compiles the project with the dependencies available.

If you immediately run cargo build again without making any changes, you won’t get any output aside from the Finished line. Cargo knows it has already downloaded and compiled the dependencies, and you haven’t changed anything about them in your Cargo.toml file. Cargo also knows that you haven’t changed anything about your code, so it doesn’t recompile that either. With nothing to do, it simply exits.

If you open up the src/main.rs file, make a trivial change, and then save it and build again, you’ll only see two lines of output:

$ cargo build
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 2.53 secs

These lines show Cargo only updates the build with your tiny change to the src/main.rs file. Your dependencies haven’t changed, so Cargo knows it can reuse what it has already downloaded and compiled for those. It just rebuilds your part of the code.

Cargo has a mechanism that ensures you can rebuild the same artifact every time you or anyone else builds your code: Cargo will use only the versions of the dependencies you specified until you indicate otherwise. For example, what happens if next week version v0.3.15 of the rand crate comes out and contains an important bug fix but also contains a regression that will break your code?

The answer to this problem is the Cargo.lock file, which was created the first time you ran cargo build and is now in your guessing_game directory. When you build a project for the first time, Cargo figures out all the versions of the dependencies that fit the criteria and then writes them to the Cargo.lock file. When you build your project in the future, Cargo will see that the Cargo.lock file exists and use the versions specified there rather than doing all the work of figuring out versions again. This lets you have a reproducible build automatically. In other words, your project will remain at 0.3.14 until you explicitly upgrade, thanks to the Cargo.lock file.

When you do want to update a crate, Cargo provides another command, update, which will ignore the Cargo.lock file and figure out all the latest versions that fit your specifications in Cargo.toml. If that works, Cargo will write those versions to the Cargo.lock file.

But by default, Cargo will only look for versions larger than 0.3.0 and smaller than 0.4.0. If the rand crate has released two new versions, 0.3.15 and 0.4.0, you would see the following if you ran cargo update:

$ cargo update
    Updating registry `https://github.com/rust-lang/crates.io-index`
    Updating rand v0.3.14 -> v0.3.15

At this point, you would also notice a change in your Cargo.lock file noting that the version of the rand crate you are now using is 0.3.15.

If you wanted to use rand version 0.4.0 or any version in the 0.4.x series, you’d have to update the Cargo.toml file to look like this instead:

[dependencies]

rand = "0.4.0"

The next time you run cargo build, Cargo will update the registry of crates available and reevaluate your rand requirements according to the new version you have specified.

There’s a lot more to say about Cargo and its ecosystem which we’ll discuss in Chapter 14, but for now, that’s all you need to know. Cargo makes it very easy to reuse libraries, so Rustaceans are able to write smaller projects that are assembled from a number of packages.

Now that you’ve added the rand crate to Cargo.toml, let’s start using rand. The next step is to update src/main.rs, as shown in Listing 2-3:

Filename: src/main.rs

extern crate rand;

use std::io;
use rand::Rng;

fn main() {
    println!("Guess the number!");

    let secret_number = rand::thread_rng().gen_range(1, 101);

    println!("The secret number is: {}", secret_number);

    println!("Please input your guess.");

    let mut guess = String::new();

    io::stdin().read_line(&mut guess)
        .expect("Failed to read line");

    println!("You guessed: {}", guess);
}

Listing 2-3: Adding code to generate a random number

First, we add a line that lets Rust know we’ll be using the rand crate as an external dependency. This also does the equivalent of calling use rand, so now we can call anything in the rand crate by placing rand:: before it.

Next, we add another use line: use rand::Rng. The Rng trait defines methods that random number generators implement, and this trait must be in scope for us to use those methods. Chapter 10 will cover traits in detail.

Also, we’re adding two more lines in the middle. The rand::thread_rng function will give us the particular random number generator that we’re going to use: one that is local to the current thread of execution and seeded by the operating system. Next, we call the gen_range method on the random number generator. This method is defined by the Rng trait that we brought into scope with the use rand::Rng statement. The gen_range method takes two numbers as arguments and generates a random number between them. It’s inclusive on the lower bound but exclusive on the upper bound, so we need to specify 1 and 101 to request a number between 1 and 100.

Note: You won’t just know which traits to use and which methods and functions to call from a crate. Instructions for using a crate are in each crate’s documentation. Another neat feature of Cargo is that you can run the cargo doc --open command, which will build documentation provided by all of your dependencies locally and open it in your browser. If you’re interested in other functionality in the rand crate, for example, run cargo doc --open and click rand in the sidebar on the left.

The second line that we added to the code prints the secret number. This is useful while we’re developing the program to be able to test it, but we’ll delete it from the final version. It’s not much of a game if the program prints the answer as soon as it starts!

Try running the program a few times:

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 2.53 secs
     Running `target/debug/guessing_game`
Guess the number!
The secret number is: 7
Please input your guess.
4
You guessed: 4
$ cargo run
     Running `target/debug/guessing_game`
Guess the number!
The secret number is: 83
Please input your guess.
5
You guessed: 5

You should get different random numbers, and they should all be numbers between 1 and 100. Great job!

Now that we have user input and a random number, we can compare them. That step is shown in Listing 2-4. Note that this code won’t compile quite yet, as we will explain.

Filename: src/main.rs

extern crate rand;

use std::io;
use std::cmp::Ordering;
use rand::Rng;

fn main() {
    // ---snip---

    println!("You guessed: {}", guess);

    match guess.cmp(&secret_number) {
        Ordering::Less => println!("Too small!"),
        Ordering::Greater => println!("Too big!"),
        Ordering::Equal => println!("You win!"),
    }
}

Listing 2-4: Handling the possible return values of comparing two numbers

The first new bit here is another use statement, bringing a type called std::cmp::Ordering into scope from the standard library. Like Result, Ordering is another enum, but the variants for Ordering are Less, Greater, and Equal. These are the three outcomes that are possible when you compare two values.

Then we add five new lines at the bottom that use the Ordering type.

The cmp method compares two values and can be called on anything that can be compared. It takes a reference to whatever you want to compare with: here it’s comparing the guess to the secret_number. Then it returns a variant of the Ordering enum we brought into scope with the use statement. We use a match expression to decide what to do next based on which variant of Ordering was returned from the call to cmp with the values in guess and secret_number.

A match expression is made up of arms. An arm consists of a pattern and the code that should be run if the value given to the beginning of the match expression fits that arm’s pattern. Rust takes the value given to match and looks through each arm’s pattern in turn. The match construct and patterns are powerful features in Rust that let you express a variety of situations your code might encounter and make sure that you handle them all. These features will be covered in detail in Chapter 6 and Chapter 18, respectively.

Let’s walk through an example of what would happen with the match expression used here. Say that the user has guessed 50 and the randomly generated secret number this time is 38. When the code compares 50 to 38, the cmp method will return Ordering::Greater, because 50 is greater than 38. The match expression gets the Ordering::Greater value and starts checking each arm’s pattern. It looks at the first arm’s pattern, Ordering::Less, and sees that the value Ordering::Greater does not match Ordering::Less, so it ignores the code in that arm and moves to the next arm. The next arm’s pattern, Ordering::Greater, does match Ordering::Greater! The associated code in that arm will execute and print Too big! to the screen. The match expression ends because it has no need to look at the last arm in this scenario.

However, the code in Listing 2-4 won’t compile yet. Let’s try it:

$ cargo build
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
error[E0308]: mismatched types
  --> src/main.rs:23:21
   |
23 |     match guess.cmp(&secret_number) {
   |                     ^^^^^^^^^^^^^^ expected struct `std::string::String`, found integral variable
   |
   = note: expected type `&std::string::String`
   = note:    found type `&{integer}`

error: aborting due to previous error
Could not compile `guessing_game`.

The core of the error states that there are mismatched types. Rust has a strong, static type system. However, it also has type inference. When we wrote let guess = String::new(), Rust was able to infer that guess should be a String and didn’t make us write the type. The secret_number, on the other hand, is a number type. A few number types can have a value between 1 and 100: i32, a 32-bit number; u32, an unsigned 32-bit number; i64, a 64-bit number; as well as others. Rust defaults to an i32, which is the type of secret_number unless you add type information elsewhere that would cause Rust to infer a different numerical type. The reason for the error is that Rust cannot compare a string and a number type.

Ultimately, we want to convert the String the program reads as input into a real number type so we can compare it numerically to the guess. We can do that by adding the following two lines to the main function body:

Filename: src/main.rs

// --snip--

    let mut guess = String::new();

    io::stdin().read_line(&mut guess)
        .expect("Failed to read line");

    let guess: u32 = guess.trim().parse()
        .expect("Please type a number!");

    println!("You guessed: {}", guess);

    match guess.cmp(&secret_number) {
        Ordering::Less => println!("Too small!"),
        Ordering::Greater => println!("Too big!"),
        Ordering::Equal => println!("You win!"),
    }
}

The two new lines are:

let guess: u32 = guess.trim().parse()
    .expect("Please type a number!");

We create a variable named guess. But wait, doesn’t the program already have a variable named guess? It does, but Rust allows us to shadow the previous value of guess with a new one. This feature is often used in situations in which you want to convert a value from one type to another type. Shadowing lets us reuse the guess variable name rather than forcing us to create two unique variables, like guess_str and guess for example. (Chapter 3 covers shadowing in more detail.)

We bind guess to the expression guess.trim().parse(). The guess in the expression refers to the original guess that was a String with the input in it. The trim method on a String instance will eliminate any whitespace at the beginning and end. Although u32 can contain only numerical characters, the user must press enter to satisfy read_line. When the user presses enter, a newline character is added to the string. For example, if the user types 5 and presses enter, guess looks like this: 5\n. The \n represents “newline,” the result of pressing enter. The trim method eliminates \n, resulting in just 5.

The parse method on strings parses a string into some kind of number. Because this method can parse a variety of number types, we need to tell Rust the exact number type we want by using let guess: u32. The colon (:) after guess tells Rust we’ll annotate the variable’s type. Rust has a few built-in number types; the u32 seen here is an unsigned, 32-bit integer. It’s a good default choice for a small positive number. You’ll learn about other number types in Chapter 3. Additionally, the u32 annotation in this example program and the comparison with secret_number means that Rust will infer that secret_number should be a u32 as well. So now the comparison will be between two values of the same type!

The call to parse could easily cause an error. If, for example, the string contained A👍%, there would be no way to convert that to a number. Because it might fail, the parse method returns a Result type, much as the read_line method does (discussed earlier in “Handling Potential Failure with the Result Type”). We’ll treat this Result the same way by using the expect method again. If parse returns an Err Result variant because it couldn’t create a number from the string, the expect call will crash the game and print the message we give it. If parse can successfully convert the string to a number, it will return the Ok variant of Result, and expect will return the number that we want from the Ok value.

Let’s run the program now!

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 0.43 secs
     Running `target/guessing_game`
Guess the number!
The secret number is: 58
Please input your guess.
  76
You guessed: 76
Too big!

Nice! Even though spaces were added before the guess, the program still figured out that the user guessed 76. Run the program a few times to verify the different behavior with different kinds of input: guess the number correctly, guess a number that is too high, and guess a number that is too low.

We have most of the game working now, but the user can make only one guess. Let’s change that by adding a loop!

The loop keyword creates an infinite loop. We’ll add that now to give users more chances at guessing the number:

Filename: src/main.rs

// --snip--

    println!("The secret number is: {}", secret_number);

    loop {
        println!("Please input your guess.");

        // --snip--

        match guess.cmp(&secret_number) {
            Ordering::Less => println!("Too small!"),
            Ordering::Greater => println!("Too big!"),
            Ordering::Equal => println!("You win!"),
        }
    }
}

As you can see, we’ve moved everything into a loop from the guess input prompt onward. Be sure to indent the lines inside the loop another four spaces each and run the program again. Notice that there is a new problem because the program is doing exactly what we told it to do: ask for another guess forever! It doesn’t seem like the user can quit!

The user could always halt the program by using the keyboard shortcut ctrl-c. But there’s another way to escape this insatiable monster, as mentioned in the parse discussion in “Comparing the Guess to the Secret Number”: if the user enters a non-number answer, the program will crash. The user can take advantage of that in order to quit, as shown here:

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
    Finished dev [unoptimized + debuginfo] target(s) in 1.50 secs
     Running `target/guessing_game`
Guess the number!
The secret number is: 59
Please input your guess.
45
You guessed: 45
Too small!
Please input your guess.
60
You guessed: 60
Too big!
Please input your guess.
59
You guessed: 59
You win!
Please input your guess.
quit
thread 'main' panicked at 'Please type a number!: ParseIntError { kind: InvalidDigit }', src/libcore/result.rs:785
note: Run with `RUST_BACKTRACE=1` for a backtrace.
error: Process didn't exit successfully: `target/debug/guess` (exit code: 101)

Typing quit actually quits the game, but so will any other non-number input. However, this is suboptimal to say the least. We want the game to automatically stop when the correct number is guessed.

Let’s program the game to quit when the user wins by adding a break statement:

Filename: src/main.rs

// --snip--

        match guess.cmp(&secret_number) {
            Ordering::Less => println!("Too small!"),
            Ordering::Greater => println!("Too big!"),
            Ordering::Equal => {
                println!("You win!");
                break;
            }
        }
    }
}

Adding the break line after You win! makes the program exit the loop when the user guesses the secret number correctly. Exiting the loop also means exiting the program, because the loop is the last part of main.

To further refine the game’s behavior, rather than crashing the program when the user inputs a non-number, let’s make the game ignore a non-number so the user can continue guessing. We can do that by altering the line where guess is converted from a String to a u32:

let guess: u32 = match guess.trim().parse() {
    Ok(num) => num,
    Err(_) => continue,
};

Switching from an expect call to a match expression is how you generally move from crashing on an error to handling the error. Remember that parse returns a Result type and Result is an enum that has the variants Ok or Err. We’re using a match expression here, as we did with the Ordering result of the cmp method.

If parse is able to successfully turn the string into a number, it will return an Ok value that contains the resulting number. That Ok value will match the first arm’s pattern, and the match expression will just return the num value that parse produced and put inside the Ok value. That number will end up right where we want it in the new guess variable we’re creating.

If parse is not able to turn the string into a number, it will return an Err value that contains more information about the error. The Err value does not match the Ok(num) pattern in the first match arm, but it does match the Err(_) pattern in the second arm. The underscore, _, is a catchall value; in this example, we’re saying we want to match all Err values, no matter what information they have inside them. So the program will execute the second arm’s code, continue, which means to go to the next iteration of the loop and ask for another guess. So effectively, the program ignores all errors that parse might encounter!

Now everything in the program should work as expected. Let’s try it:

$ cargo run
   Compiling guessing_game v0.1.0 (file:///projects/guessing_game)
     Running `target/guessing_game`
Guess the number!
The secret number is: 61
Please input your guess.
10
You guessed: 10
Too small!
Please input your guess.
99
You guessed: 99
Too big!
Please input your guess.
foo
Please input your guess.
61
You guessed: 61
You win!

Awesome! With one tiny final tweak, we will finish the guessing game. Recall that the program is still printing the secret number. That worked well for testing, but it ruins the game. Let’s delete the println! that outputs the secret number. Listing 2-5 shows the final code:

Filename: src/main.rs

extern crate rand;

use std::io;
use std::cmp::Ordering;
use rand::Rng;

fn main() {
    println!("Guess the number!");

    let secret_number = rand::thread_rng().gen_range(1, 101);

    loop {
        println!("Please input your guess.");

        let mut guess = String::new();

        io::stdin().read_line(&mut guess)
            .expect("Failed to read line");

        let guess: u32 = match guess.trim().parse() {
            Ok(num) => num,
            Err(_) => continue,
        };

        println!("You guessed: {}", guess);

        match guess.cmp(&secret_number) {
            Ordering::Less => println!("Too small!"),
            Ordering::Greater => println!("Too big!"),
            Ordering::Equal => {
                println!("You win!");
                break;
            }
        }
    }
}

Listing 2-5: Complete guessing game code

At this point, you’ve successfully built the guessing game! Congratulations!

This project was a hands-on way to introduce you to many new Rust concepts: let, match, methods, associated functions, the use of external crates, and more. In the next few chapters, you’ll learn about these concepts in more detail. Chapter 3 covers concepts that most programming languages have, such as variables, data types, and functions, and shows how to use them in Rust. Chapter 4 explores ownership, a feature that makes Rust different from other languages. Chapter 5 discusses structs and method syntax, and Chapter 6 explains how enums work.

This chapter covers concepts that appear in almost every programming language and how they work in Rust. Many programming languages have much in common at their core. None of the concepts presented in this chapter are unique to Rust, but we'll discuss them in the context of Rust and explain the conventions around using these concepts.

Specifically, you’ll learn about variables, basic types, functions, comments, and control flow. These foundations will be in every Rust program, and learning them early will give you a strong core to start from.

Keywords

The Rust language has a set of keywords that are reserved for use by the language only, much as in other languages. Keep in mind that you cannot use these words as names of variables or functions. Most of the keywords have special meanings, and you’ll be using them to do various tasks in your Rust programs; a few have no current functionality associated with them but have been reserved for functionality that might be added to Rust in the future. You can find a list of the keywords in Appendix A.

As mentioned in Chapter 2, by default variables are immutable. This is one of many nudges Rust gives you to write your code in a way that takes advantage of the safety and easy concurrency that Rust offers. However, you still have the option to make your variables mutable. Let’s explore how and why Rust encourages you to favor immutability and why sometimes you might want to opt out.

When a variable is immutable, once a value is bound to a name, you can’t change that value. To illustrate this, let’s generate a new project called variables in your projects directory by using cargo new --bin variables.

Then, in your new variables directory, open src/main.rs and replace its code with the following code that won’t compile just yet:

Filename: src/main.rs

fn main() {
    let x = 5;
    println!("The value of x is: {}", x);
    x = 6;
    println!("The value of x is: {}", x);
}

Save and run the program using cargo run. You should receive an error message, as shown in this output:

error[E0384]: cannot assign twice to immutable variable `x`
 --> src/main.rs:4:5
  |
2 |     let x = 5;
  |         - first assignment to `x`
3 |     println!("The value of x is: {}", x);
4 |     x = 6;
  |     ^^^^^ cannot assign twice to immutable variable

This example shows how the compiler helps you find errors in your programs. Even though compiler errors can be frustrating, they only mean your program isn’t safely doing what you want it to do yet; they do not mean that you’re not a good programmer! Experienced Rustaceans still get compiler errors.

The error indicates that the cause of the error is that you cannot assign twice to immutable variable x, because you tried to assign a second value to the immutable x variable.

It’s important that we get compile-time errors when we attempt to change a value that we previously designated as immutable because this very situation can lead to bugs. If one part of our code operates on the assumption that a value will never change and another part of our code changes that value, it’s possible that the first part of the code won’t do what it was designed to do. The cause of this kind of bug can be difficult to track down after the fact, especially when the second piece of code changes the value only sometimes.

In Rust, the compiler guarantees that when you state that a value won’t change, it really won’t change. That means that when you’re reading and writing code, you don’t have to keep track of how and where a value might change. Your code is thus easier to reason through.

But mutability can be very useful. Variables are immutable only by default; as you did in Chapter 2, you can make them mutable by adding mut in front of the variable name. In addition to allowing this value to change, mut conveys intent to future readers of the code by indicating that other parts of the code will be changing this variable value.

For example, let’s change src/main.rs to the following:

Filename: src/main.rs

fn main() {
    let mut x = 5;
    println!("The value of x is: {}", x);
    x = 6;
    println!("The value of x is: {}", x);
}

When we run the program now, we get this:

$ cargo run
   Compiling variables v0.1.0 (file:///projects/variables)
    Finished dev [unoptimized + debuginfo] target(s) in 0.30 secs
     Running `target/debug/variables`
The value of x is: 5
The value of x is: 6

We’re allowed to change the value that x binds to from 5 to 6 when mut is used. In some cases, you’ll want to make a variable mutable because it makes the code more convenient to write than if it had only immutable variables.

There are multiple trade-offs to consider in addition to the prevention of bugs. For example, in cases where you’re using large data structures, mutating an instance in place may be faster than copying and returning newly allocated instances. With smaller data structures, creating new instances and writing in a more functional programming style may be easier to think through, so lower performance might be a worthwhile penalty for gaining that clarity.

Being unable to change the value of a variable might have reminded you of another programming concept that most other languages have: constants. Like immutable variables, constants are values that are bound to a name and are not allowed to change, but there are a few differences between constants and variables.

First, you aren’t allowed to use mut with constants. Constants aren’t just immutable by default—they’re always immutable.

You declare constants using the const keyword instead of the let keyword, and the type of the value must be annotated. We’re about to cover types and type annotations in the next section, “Data Types,” so don’t worry about the details right now. Just know that you must always annotate the type.

Constants can be declared in any scope, including the global scope, which makes them useful for values that many parts of code need to know about.

The last difference is that constants may be set only to a constant expression, not the result of a function call or any other value that could only be computed at runtime.

Here’s an example of a constant declaration where the constant’s name is MAX_POINTS and its value is set to 100,000. (Rust’s constant naming convention is to use all uppercase with underscores between words):

# #![allow(unused_variables)]
#fn main() {
const MAX_POINTS: u32 = 100_000;
#}

Constants are valid for the entire time a program runs, within the scope they were declared in, making them a useful choice for values in your application domain that multiple parts of the program might need to know about, such as the maximum number of points any player of a game is allowed to earn or the speed of light.

Naming hardcoded values used throughout your program as constants is useful in conveying the meaning of that value to future maintainers of the code. It also helps to have only one place in your code you would need to change if the hardcoded value needed to be updated in the future.

As you saw in the “Comparing the Guess to the Secret Number” section in Chapter 2, you can declare a new variable with the same name as a previous variable, and the new variable shadows the previous variable. Rustaceans say that the first variable is shadowed by the second, which means that the second variable’s value is what appears when the variable is used. We can shadow a variable by using the same variable’s name and repeating the use of the let keyword as follows:

Filename: src/main.rs

fn main() {
    let x = 5;

    let x = x + 1;

    let x = x * 2;

    println!("The value of x is: {}", x);
}

This program first binds x to a value of 5. Then it shadows x by repeating let x =, taking the original value and adding 1 so the value of x is then 6. The third let statement also shadows x, multiplying the previous value by 2 to give x a final value of 12. When we run this program, it will output the following:

$ cargo run
   Compiling variables v0.1.0 (file:///projects/variables)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/variables`
The value of x is: 12

Shadowing is different than marking a variable as mut, because we’ll get a compile-time error if we accidentally try to reassign to this variable without using the let keyword. By using let, we can perform a few transformations on a value but have the variable be immutable after those transformations have been completed.

The other difference between mut and shadowing is that because we’re effectively creating a new variable when we use the let keyword again, we can change the type of the value but reuse the same name. For example, say our program asks a user to show how many spaces they want between some text by inputting space characters, but we really want to store that input as a number:

# #![allow(unused_variables)]
#fn main() {
let spaces = "   ";
let spaces = spaces.len();
#}

This construct is allowed because the first spaces variable is a string type and the second spaces variable, which is a brand-new variable that happens to have the same name as the first one, is a number type. Shadowing thus spares us from having to come up with different names, such as spaces_str and spaces_num; instead, we can reuse the simpler spaces name. However, if we try to use mut for this, as shown here, we’ll get a compile-time error:

let mut spaces = "   ";
spaces = spaces.len();

The error says we’re not allowed to mutate a variable’s type:

error[E0308]: mismatched types
 --> src/main.rs:3:14
  |
3 |     spaces = spaces.len();
  |              ^^^^^^^^^^^^ expected &str, found usize
  |
  = note: expected type `&str`
             found type `usize`

Now that we’ve explored how variables work, let’s look at more data types they can have.

Every value in Rust is of a certain data type, which tells Rust what kind of data is being specified so it knows how to work with that data. We’ll look at two data type subsets: scalar and compound.

Keep in mind that Rust is a statically typed language, which means that it must know the types of all variables at compile time. The compiler can usually infer what type we want to use based on the value and how we use it. In cases when many types are possible, such as when we converted a String to a numeric type using parse in the “Comparing the Guess to the Secret Number” section in Chapter 2, we must add a type annotation, like this:

# #![allow(unused_variables)]
#fn main() {
let guess: u32 = "42".parse().expect("Not a number!");
#}

If we don’t add the type annotation here, Rust will display the following error, which means the compiler needs more information from us to know which type we want to use:

error[E0282]: type annotations needed
 --> src/main.rs:2:9
  |
2 |     let guess = "42".parse().expect("Not a number!");
  |         ^^^^^
  |         |
  |         cannot infer type for `_`
  |         consider giving `guess` a type

You’ll see different type annotations for other data types.

A scalar type represents a single value. Rust has four primary scalar types: integers, floating-point numbers, Booleans, and characters. You may recognize these from other programming languages. Let’s jump into how they work in Rust.

An integer is a number without a fractional component. We used one integer type in Chapter 2, the u32 type. This type declaration indicates that the value it’s associated with should be an unsigned integer (signed integer types start with i, instead of u) that takes up 32 bits of space. Table 3-1 shows the built-in integer types in Rust. Each variant in the Signed and Unsigned columns (for example, i16) can be used to declare the type of an integer value.

Table 3-1: Integer Types in Rust

Each variant can be either signed or unsigned and has an explicit size. Signed and unsigned refer to whether it’s possible for the number to be negative or positive—in other words, whether the number needs to have a sign with it (signed) or whether it will only ever be positive and can therefore be represented without a sign (unsigned). It’s like writing numbers on paper: when the sign matters, a number is shown with a plus sign or a minus sign; however, when it’s safe to assume the number is positive, it’s shown with no sign. Signed numbers are stored using two’s complement representation (if you’re unsure what this is, you can search for it online; an explanation is outside the scope of this book).

Each signed variant can store numbers from -(2^{n - 1}) to 2^{n - 1} - 1 inclusive, where n is the number of bits that variant uses. So an i8 can store numbers from -(2⁷) to 2⁷ - 1, which equals -128 to 127. Unsigned variants can store numbers from 0 to 2ⁿ - 1, so a u8 can store numbers from 0 to 2⁸ - 1, which equals 0 to 255.

Additionally, the isize and usize types depend on the kind of computer your program is running on: 64 bits if you’re on a 64-bit architecture and 32 bits if you’re on a 32-bit architecture.

You can write integer literals in any of the forms shown in Table 3-2. Note that all number literals except the byte literal allow a type suffix, such as 57u8, and _ as a visual separator, such as 1_000.

Table 3-2: Integer Literals in Rust

So how do you know which type of integer to use? If you’re unsure, Rust’s defaults are generally good choices, and integer types default to i32: this type is generally the fastest, even on 64-bit systems. The primary situation in which you’d use isize or usize is when indexing some sort of collection.

Rust also has two primitive types for floating-point numbers, which are numbers with decimal points. Rust’s floating-point types are f32 and f64, which are 32 bits and 64 bits in size, respectively. The default type is f64 because on modern CPUs it’s roughly the same speed as f32 but is capable of more precision.

Here’s an example that shows floating-point numbers in action:

Filename: src/main.rs

fn main() {
    let x = 2.0; // f64

    let y: f32 = 3.0; // f32
}

Floating-point numbers are represented according to the IEEE-754 standard. The f32 type is a single-precision float, and f64 has double precision.

Rust supports the basic mathematical operations you’d expect for all of the number types: addition, subtraction, multiplication, division, and remainder. The following code shows how you’d use each one in a let statement:

Filename: src/main.rs

fn main() {
    // addition
    let sum = 5 + 10;

    // subtraction
    let difference = 95.5 - 4.3;

    // multiplication
    let product = 4 * 30;

    // division
    let quotient = 56.7 / 32.2;

    // remainder
    let remainder = 43 % 5;
}

Each expression in these statements uses a mathematical operator and evaluates to a single value, which is then bound to a variable. Appendix B contains a list of all operators that Rust provides.

As in most other programming languages, a Boolean type in Rust has two possible values: true and false. The Boolean type in Rust is specified using bool. For example:

Filename: src/main.rs

fn main() {
    let t = true;

    let f: bool = false; // with explicit type annotation
}

The main way to consume Boolean values is through conditionals, such as an if expression. We’ll cover how if expressions work in Rust in the “Control Flow” section.

So far we’ve worked only with numbers, but Rust supports letters too. Rust’s char type is the language’s most primitive alphabetic type, and the following code shows one way to use it. (Note that the char type is specified with single quotes, as opposed to strings, which use double quotes.)

Filename: src/main.rs

fn main() {
    let c = 'z';
    let z = 'ℤ';
    let heart_eyed_cat = '😻';
}

Rust’s char type represents a Unicode Scalar Value, which means it can represent a lot more than just ASCII. Accented letters; Chinese, Japanese, and Korean ideographs; emoji; and zero-width spaces are all valid char values in Rust. Unicode Scalar Values range from U+0000 to U+D7FF and U+E000 to U+10FFFF inclusive. However, a “character” isn’t really a concept in Unicode, so your human intuition for what a “character” is may not match up with what a char is in Rust. We’ll discuss this topic in detail in “Strings” in Chapter 8.

Compound types can group multiple values into one type. Rust has two primitive compound types: tuples and arrays.

A tuple is a general way of grouping together some number of other values with a variety of types into one compound type.

We create a tuple by writing a comma-separated list of values inside parentheses. Each position in the tuple has a type, and the types of the different values in the tuple don’t have to be the same. We’ve added optional type annotations in this example:

Filename: src/main.rs

fn main() {
    let tup: (i32, f64, u8) = (500, 6.4, 1);
}

The variable tup binds to the entire tuple, because a tuple is considered a single compound element. To get the individual values out of a tuple, we can use pattern matching to destructure a tuple value, like this:

Filename: src/main.rs

fn main() {
    let tup = (500, 6.4, 1);

    let (x, y, z) = tup;

    println!("The value of y is: {}", y);
}

This program first creates a tuple and binds it to the variable tup. It then uses a pattern with let to take tup and turn it into three separate variables, x, y, and z. This is called destructuring, because it breaks the single tuple into three parts. Finally, the program prints the value of y, which is 6.4.

In addition to destructuring through pattern matching, we can access a tuple element directly by using a period (.) followed by the index of the value we want to access. For example:

Filename: src/main.rs

fn main() {
    let x: (i32, f64, u8) = (500, 6.4, 1);

    let five_hundred = x.0;

    let six_point_four = x.1;

    let one = x.2;
}

This program creates a tuple, x, and then makes new variables for each element by using their index. As with most programming languages, the first index in a tuple is 0.

Another way to have a collection of multiple values is with an array. Unlike a tuple, every element of an array must have the same type. Arrays in Rust are different from arrays in some other languages because arrays in Rust have a fixed length: once declared, they cannot grow or shrink in size.

In Rust, the values going into an array are written as a comma-separated list inside square brackets:

Filename: src/main.rs

fn main() {
    let a = [1, 2, 3, 4, 5];
}

Arrays are useful when you want your data allocated on the stack rather than the heap (we will discuss the stack and the heap more in Chapter 4), or when you want to ensure you always have a fixed number of elements. An array isn’t as flexible as the vector type, though. A vector is a similar collection type provided by the standard library that is allowed to grow or shrink in size. If you’re unsure whether to use an array or a vector, you should probably use a vector. Chapter 8 discusses vectors in more detail.

An example of when you might want to use an array rather than a vector is in a program that needs to know the names of the months of the year. It’s very unlikely that such a program will need to add or remove months, so you can use an array because you know it will always contain 12 items:

# #![allow(unused_variables)]
#fn main() {
let months = ["January", "February", "March", "April", "May", "June", "July",
              "August", "September", "October", "November", "December"];
#}

An array is a single chunk of memory allocated on the stack. You can access elements of an array using indexing, like this:

Filename: src/main.rs

fn main() {
    let a = [1, 2, 3, 4, 5];

    let first = a[0];
    let second = a[1];
}

In this example, the variable named first will get the value 1, because that is the value at index [0] in the array. The variable named second will get the value 2 from index [1] in the array.

What happens if you try to access an element of an array that is past the end of the array? Say you change the example to the following code, which will compile but exit with an error when it runs:

Filename: src/main.rs

fn main() {
    let a = [1, 2, 3, 4, 5];
    let index = 10;

    let element = a[index];

    println!("The value of element is: {}", element);
}

Running this code using cargo run produces the following result:

$ cargo run
   Compiling arrays v0.1.0 (file:///projects/arrays)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/arrays`
thread '<main>' panicked at 'index out of bounds: the len is 5 but the index is
 10', src/main.rs:6
note: Run with `RUST_BACKTRACE=1` for a backtrace.

The compilation didn’t produce any errors, but the program resulted in a runtime error and didn’t exit successfully. When you attempt to access an element using indexing, Rust will check that the index you’ve specified is less than the array length. If the index is greater than the length, Rust will panic, which is the term Rust uses when a program exits with an error.

This is the first example of Rust’s safety principles in action. In many low-level languages, this kind of check is not done, and when you provide an incorrect index, invalid memory can be accessed. Rust protects you against this kind of error by immediately exiting instead of allowing the memory access and continuing. Chapter 9 discusses more of Rust’s error handling.

Functions are pervasive in Rust code. You’ve already seen one of the most important functions in the language: the main function, which is the entry point of many programs. You’ve also seen the fn keyword, which allows you to declare new functions.

Rust code uses snake case as the conventional style for function and variable names. In snake case, all letters are lowercase and underscores separate words. Here’s a program that contains an example function definition:

Filename: src/main.rs

fn main() {
    println!("Hello, world!");

    another_function();
}

fn another_function() {
    println!("Another function.");
}

Function definitions in Rust start with fn and have a set of parentheses after the function name. The curly brackets tell the compiler where the function body begins and ends.

We can call any function we’ve defined by entering its name followed by a set of parentheses. Because another_function is defined in the program, it can be called from inside the main function. Note that we defined another_function after the main function in the source code; we could have defined it before as well. Rust doesn’t care where you define your functions, only that they’re defined somewhere.

Let’s start a new binary project named functions to explore functions further. Place the another_function example in src/main.rs and run it. You should see the following output:

$ cargo run
   Compiling functions v0.1.0 (file:///projects/functions)
    Finished dev [unoptimized + debuginfo] target(s) in 0.28 secs
     Running `target/debug/functions`
Hello, world!
Another function.

The lines execute in the order in which they appear in the main function. First, the “Hello, world!” message prints, and then another_function is called and its message is printed.

Functions can also be defined to have parameters, which are special variables that are part of a function’s signature. When a function has parameters, you can provide it with concrete values for those parameters. Technically, the concrete values are called arguments, but in casual conversation, people tend to use the words parameter and argument interchangeably for either the variables in a function’s definition or the concrete values passed in when you call a function.

The following rewritten version of another_function shows what parameters look like in Rust:

Filename: src/main.rs

fn main() {
    another_function(5);
}

fn another_function(x: i32) {
    println!("The value of x is: {}", x);
}

Try running this program; you should get the following output:

$ cargo run
   Compiling functions v0.1.0 (file:///projects/functions)
    Finished dev [unoptimized + debuginfo] target(s) in 1.21 secs
     Running `target/debug/functions`
The value of x is: 5

The declaration of another_function has one parameter named x. The type of x is specified as i32. When 5 is passed to another_function, the println! macro puts 5 where the pair of curly brackets were in the format string.

In function signatures, you must declare the type of each parameter. This is a deliberate decision in Rust’s design: requiring type annotations in function definitions means the compiler almost never needs you to use them elsewhere in the code to figure out what you mean.

When you want a function to have multiple parameters, separate the parameter declarations with commas, like this:

Filename: src/main.rs

fn main() {
    another_function(5, 6);
}

fn another_function(x: i32, y: i32) {
    println!("The value of x is: {}", x);
    println!("The value of y is: {}", y);
}

This example creates a function with two parameters, both of which are i32 types. The function then prints the values in both of its parameters. Note that function parameters don’t all need to be the same type, they just happen to be in this example.

Let’s try running this code. Replace the program currently in your functions project’s src/main.rs file with the preceding example and run it using cargo run:

$ cargo run
   Compiling functions v0.1.0 (file:///projects/functions)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/functions`
The value of x is: 5
The value of y is: 6

Because we called the function with 5 as the value for x and 6 is passed as the value for y, the two strings are printed with these values.

Function bodies are made up of a series of statements optionally ending in an expression. So far, we’ve only covered functions without an ending expression, but you have seen an expression as part of statements. Because Rust is an expression-based language, this is an important distinction to understand. Other languages don’t have the same distinctions, so let’s look at what statements and expressions are and how their differences affect the bodies of functions.

We’ve actually already used statements and expressions. Statements are instructions that perform some action and do not return a value. Expressions evaluate to a resulting value. Let’s look at some examples.

Creating a variable and assigning a value to it with the let keyword is a statement. In Listing 3-1, let y = 6; is a statement:

Filename: src/main.rs

fn main() {
    let y = 6;
}

Listing 3-1: A main function declaration containing one statement

Function definitions are also statements; the entire preceding example is a statement in itself.

Statements do not return values. Therefore, you can’t assign a let statement to another variable, as the following code tries to do; you’ll get an error:

Filename: src/main.rs

fn main() {
    let x = (let y = 6);
}

When you run this program, the error you’ll get looks like this:

$ cargo run
   Compiling functions v0.1.0 (file:///projects/functions)
error: expected expression, found statement (`let`)
 --> src/main.rs:2:14
  |
2 |     let x = (let y = 6);
  |              ^^^
  |
  = note: variable declaration using `let` is a statement

The let y = 6 statement does not return a value, so there isn’t anything for x to bind to. This is different from what happens in other languages, such as C and Ruby, where the assignment returns the value of the assignment. In those languages, you can write x = y = 6 and have both x and y have the value 6; that is not the case in Rust.

Expressions evaluate to something and make up most of the rest of the code that you’ll write in Rust. Consider a simple math operation, such as 5 + 6, which is an expression that evaluates to the value 11. Expressions can be part of statements: in Listing 3-1, the 6 in the statement let y = 6; is an expression that evaluates to the value 6. Calling a function is an expression. Calling a macro is an expression. The block that we use to create new scopes, {}, is an expression, for example:

Filename: src/main.rs

fn main() {
    let x = 5;

    let y = {
        let x = 3;
        x + 1
    };

    println!("The value of y is: {}", y);
}

This expression:

{
    let x = 3;
    x + 1
}

is a block that, in this case, evaluates to 4. That value gets bound to y as part of the let statement. Note the x + 1 line without a semicolon at the end, which is unlike most of the lines you’ve seen so far. Expressions do not include ending semicolons. If you add a semicolon to the end of an expression, you turn it into a statement, which will then not return a value. Keep this in mind as you explore function return values and expressions next.

Functions can return values to the code that calls them. We don’t name return values, but we do declare their type after an arrow (->). In Rust, the return value of the function is synonymous with the value of the final expression in the block of the body of a function. You can return early from a function by using the return keyword and specifying a value, but most functions return the last expression implicitly. Here’s an example of a function that returns a value:

Filename: src/main.rs

fn five() -> i32 {
    5
}

fn main() {
    let x = five();

    println!("The value of x is: {}", x);
}

There are no function calls, macros, or even let statements in the five function—just the number 5 by itself. That’s a perfectly valid function in Rust. Note that the function’s return type is specified, too, as -> i32. Try running this code; the output should look like this:

$ cargo run
   Compiling functions v0.1.0 (file:///projects/functions)
    Finished dev [unoptimized + debuginfo] target(s) in 0.30 secs
     Running `target/debug/functions`
The value of x is: 5

The 5 in five is the function’s return value, which is why the return type is i32. Let’s examine this in more detail. There are two important bits: first, the line let x = five(); shows that we’re using the return value of a function to initialize a variable. Because the function five returns a 5, that line is the same as the following:

# #![allow(unused_variables)]
#fn main() {
let x = 5;
#}

Second, the five function has no parameters and defines the type of the return value, but the body of the function is a lonely 5 with no semicolon because it’s an expression whose value we want to return.

Let’s look at another example:

Filename: src/main.rs

fn main() {
    let x = plus_one(5);

    println!("The value of x is: {}", x);
}

fn plus_one(x: i32) -> i32 {
    x + 1
}

Running this code will print The value of x is: 6. But if we place a semicolon at the end of the line containing x + 1, changing it from an expression to a statement, we’ll get an error.

Filename: src/main.rs

fn main() {
    let x = plus_one(5);

    println!("The value of x is: {}", x);
}

fn plus_one(x: i32) -> i32 {
    x + 1;
}

Running this code produces an error, as follows:

error[E0308]: mismatched types
 --> src/main.rs:7:28
  |
7 |   fn plus_one(x: i32) -> i32 {
  |  ____________________________^
8 | |     x + 1;
  | |          - help: consider removing this semicolon
9 | | }
  | |_^ expected i32, found ()
  |
  = note: expected type `i32`
             found type `()`

The main error message, “mismatched types,” reveals the core issue with this code. The definition of the function plus_one says that it will return an i32, but statements don’t evaluate to a value, which is expressed by (), the empty tuple. Therefore, nothing is returned, which contradicts the function definition and results in an error. In this output, Rust provides a message to possibly help rectify this issue: it suggests removing the semicolon, which would fix the error.

All programmers strive to make their code easy to understand, but sometimes extra explanation is warranted. In these cases, programmers leave notes, or comments, in their source code that the compiler will ignore but people reading the source code may find useful.

Here’s a simple comment:

# #![allow(unused_variables)]
#fn main() {
// Hello, world.
#}

In Rust, comments must start with two slashes and continue until the end of the line. For comments that extend beyond a single line, you’ll need to include // on each line, like this:

# #![allow(unused_variables)]
#fn main() {
// So we’re doing something complicated here, long enough that we need
// multiple lines of comments to do it! Whew! Hopefully, this comment will
// explain what’s going on.
#}

Comments can also be placed at the end of lines containing code:

Filename: src/main.rs

fn main() {
    let lucky_number = 7; // I’m feeling lucky today.
}

But you’ll more often see them used in this format, with the comment on a separate line above the code it’s annotating:

Filename: src/main.rs

fn main() {
    // I’m feeling lucky today.
    let lucky_number = 7;
}

Rust also has another kind of comment, documentation comments, which we’ll discuss in Chapter 14.

Deciding whether or not to run some code depending on if a condition is true and deciding to run some code repeatedly while a condition is true are basic building blocks in most programming languages. The most common constructs that let you control the flow of execution of Rust code are if expressions and loops.

An if expression allows you to branch your code depending on conditions. You provide a condition and then state, “If this condition is met, run this block of code. If the condition is not met, do not run this block of code.”

Create a new project called branches in your projects directory to explore the if expression. In the src/main.rs file, input the following:

Filename: src/main.rs

fn main() {
    let number = 3;

    if number < 5 {
        println!("condition was true");
    } else {
        println!("condition was false");
    }
}

All if expressions start with the keyword if, which is followed by a condition. In this case, the condition checks whether or not the variable number has a value less than 5. The block of code we want to execute if the condition is true is placed immediately after the condition inside curly brackets. Blocks of code associated with the conditions in if expressions are sometimes called arms, just like the arms in match expressions that we discussed in the “Comparing the Guess to the Secret Number” section of Chapter 2.

Optionally, we can also include an else expression, which we chose to do here, to give the program an alternative block of code to execute should the condition evaluate to false. If you don’t provide an else expression and the condition is false, the program will just skip the if block and move on to the next bit of code.

Try running this code; you should see the following output:

$ cargo run
   Compiling branches v0.1.0 (file:///projects/branches)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/branches`
condition was true

Let’s try changing the value of number to a value that makes the condition false to see what happens:

let number = 7;

Run the program again, and look at the output:

$ cargo run
   Compiling branches v0.1.0 (file:///projects/branches)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/branches`
condition was false

It’s also worth noting that the condition in this code must be a bool. If the condition isn’t a bool, we’ll get an error. For example:

Filename: src/main.rs

fn main() {
    let number = 3;

    if number {
        println!("number was three");
    }
}

The if condition evaluates to a value of 3 this time, and Rust throws an error:

error[E0308]: mismatched types
 --> src/main.rs:4:8
  |
4 |     if number {
  |        ^^^^^^ expected bool, found integral variable
  |
  = note: expected type `bool`
             found type `{integer}`

The error indicates that Rust expected a bool but got an integer. Unlike languages such as Ruby and JavaScript, Rust will not automatically try to convert non-Boolean types to a Boolean. You must be explicit and always provide if with a Boolean as its condition. If we want the if code block to run only when a number is not equal to 0, for example, we can change the if expression to the following:

Filename: src/main.rs

fn main() {
    let number = 3;

    if number != 0 {
        println!("number was something other than zero");
    }
}

Running this code will print number was something other than zero.

You can have multiple conditions by combining if and else in an else if expression. For example:

Filename: src/main.rs

fn main() {
    let number = 6;

    if number % 4 == 0 {
        println!("number is divisible by 4");
    } else if number % 3 == 0 {
        println!("number is divisible by 3");
    } else if number % 2 == 0 {
        println!("number is divisible by 2");
    } else {
        println!("number is not divisible by 4, 3, or 2");
    }
}

This program has four possible paths it can take. After running it, you should see the following output:

$ cargo run
   Compiling branches v0.1.0 (file:///projects/branches)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running `target/debug/branches`
number is divisible by 3

When this program executes, it checks each if expression in turn and executes the first body for which the condition holds true. Note that even though 6 is divisible by 2, we don’t see the output number is divisible by 2, nor do we see the number is not divisible by 4, 3, or 2 text from the else block. That’s because Rust only executes the block for the first true condition, and once it finds one, it doesn’t even check the rest.

Using too many else if expressions can clutter your code, so if you have more than one, you might want to refactor your code. Chapter 6 describes a powerful Rust branching construct called match for these cases.

Because if is an expression, we can use it on the right side of a let statement, as in Listing 3-2:

Filename: src/main.rs

fn main() {
    let condition = true;
    let number = if condition {
        5
    } else {
        6
    };

    println!("The value of number is: {}", number);
}

Listing 3-2: Assigning the result of an if expression to a variable

The number variable will be bound to a value based on the outcome of the if expression. Run this code to see what happens:

$ cargo run
   Compiling branches v0.1.0 (file:///projects/branches)
    Finished dev [unoptimized + debuginfo] target(s) in 0.30 secs
     Running `target/debug/branches`
The value of number is: 5

Remember that blocks of code evaluate to the last expression in them, and numbers by themselves are also expressions. In this case, the value of the whole if expression depends on which block of code executes. This means the values that have the potential to be results from each arm of the if must be the same type; in Listing 3-2, the results of both the if arm and the else arm were i32 integers. If the types are mismatched, as in the following example, we’ll get an error:

Filename: src/main.rs

fn main() {
    let condition = true;

    let number = if condition {
        5
    } else {
        "six"
    };

    println!("The value of number is: {}", number);
}

When we try to run this code, we’ll get an error. The if and else arms have value types that are incompatible, and Rust indicates exactly where to find the problem in the program:

error[E0308]: if and else have incompatible types
 --> src/main.rs:4:18
  |
4 |       let number = if condition {
  |  __________________^
5 | |         5
6 | |     } else {
7 | |         "six"
8 | |     };
  | |_____^ expected integral variable, found &str
  |
  = note: expected type `{integer}`
             found type `&str`

The expression in the if block evaluates to an integer, and the expression in the else block evaluates to a string. This won’t work because variables must have a single type. Rust needs to know at compile time what type the number variable is, definitively, so it can verify at compile time that its type is valid everywhere we use number. Rust wouldn’t be able to do that if the type of number was only determined at runtime; the compiler would be more complex and would make fewer guarantees about the code if it had to keep track of multiple hypothetical types for any variable.

It’s often useful to execute a block of code more than once. For this task, Rust provides several loops. A loop runs through the code inside the loop body to the end and then starts immediately back at the beginning. To experiment with loops, let’s make a new project called loops.

Rust has three kinds of loops: loop, while, and for. Let’s try each one.

The loop keyword tells Rust to execute a block of code over and over again forever or until you explicitly tell it to stop.

As an example, change the src/main.rs file in your loops directory to look like this:

Filename: src/main.rs

fn main() {
    loop {
        println!("again!");
    }
}

When we run this program, we’ll see again! printed over and over continuously until we stop the program manually. Most terminals support a keyboard shortcut, ctrl-c, to halt a program that is stuck in a continual loop. Give it a try:

$ cargo run
   Compiling loops v0.1.0 (file:///projects/loops)
    Finished dev [unoptimized + debuginfo] target(s) in 0.29 secs
     Running `target/debug/loops`
again!
again!
again!
again!
^Cagain!

The symbol ^C represents where you pressed ctrl-c . You may or may not see the word again! printed after the ^C, depending on where the code was in the loop when it received the halt signal.

Fortunately, Rust provides another, more reliable way to break out of a loop. You can place the break keyword within the loop to tell the program when to stop executing the loop. Recall that we did this in the guessing game in the “Quitting After a Correct Guess” section of Chapter 2 to exit the program when the user won the game by guessing the correct number.

It’s often useful for a program to evaluate a condition within a loop. While the condition is true, the loop runs. When the condition ceases to be true, the program calls break, stopping the loop. This loop type could be implemented using a combination of loop, if, else, and break; you could try that now in a program, if you’d like.

However, this pattern is so common that Rust has a built-in language construct for it, called a while loop. Listing 3-3 uses while: the program loops three times, counting down each time, and then, after the loop, it prints another message and exits.

Filename: src/main.rs

fn main() {
    let mut number = 3;

    while number != 0 {
        println!("{}!", number);

        number = number - 1;
    }

    println!("LIFTOFF!!!");
}

Listing 3-3: Using a while loop to run code while a condition holds true

This construct eliminates a lot of nesting that would be necessary if you used loop, if, else, and break, and it’s clearer. While a condition holds true, the code runs; otherwise, it exits the loop.

You could use the while construct to loop over the elements of a collection, such as an array. For example, let’s look at Listing 3-4:

Filename: src/main.rs

fn main() {
    let a = [10, 20, 30, 40, 50];
    let mut index = 0;

    while index < 5 {
        println!("the value is: {}", a[index]);

        index = index + 1;
    }
}

Listing 3-4: Looping through each element of a collection using a while loop

Here, the code counts up through the elements in the array. It starts at index 0, and then loops until it reaches the final index in the array (that is, when index < 5 is no longer true). Running this code will print every element in the array:

$ cargo run
   Compiling loops v0.1.0 (file:///projects/loops)
    Finished dev [unoptimized + debuginfo] target(s) in 0.32 secs
     Running `target/debug/loops`
the value is: 10
the value is: 20
the value is: 30
the value is: 40
the value is: 50

All five array values appear in the terminal, as expected. Even though index will reach a value of 5 at some point, the loop stops executing before trying to fetch a sixth value from the array.

But this approach is error prone; we could cause the program to panic if the index length is incorrect. It’s also slow, because the compiler adds runtime code to perform the conditional check on every element on every iteration through the loop.

As a more concise alternative, you can use a for loop and execute some code for each item in a collection. A for loop looks like this code in Listing 3-5:

Filename: src/main.rs

fn main() {
    let a = [10, 20, 30, 40, 50];

    for element in a.iter() {
        println!("the value is: {}", element);
    }
}

Listing 3-5: Looping through each element of a collection using a for loop

When we run this code, we’ll see the same output as in Listing 3-4. More importantly, we’ve now increased the safety of the code and eliminated the chance of bugs that might result from going beyond the end of the array or not going far enough and missing some items.

For example, in the code in Listing 3-4, if you removed an item from the a array but forgot to update the condition to while index < 4, the code would panic. Using the for loop, you wouldn’t need to remember to change any other code if you changed the number of values in the array.

The safety and conciseness of for loops make them the most commonly used loop construct in Rust. Even in situations in which you want to run some code a certain number of times, as in the countdown example that used a while loop in Listing 3-3, most Rustaceans would use a for loop. The way to do that would be to use a Range, which is a type provided by the standard library that generates all numbers in sequence starting from one number and ending before another number.

Here’s what the countdown would look like using a for loop and another method we’ve not yet talked about, rev, to reverse the range:

Filename: src/main.rs

fn main() {
    for number in (1..4).rev() {
        println!("{}!", number);
    }
    println!("LIFTOFF!!!");
}

This code is a bit nicer, isn’t it?

You made it! That was a sizable chapter: you learned about variables, scalar and compound data types, functions, comments, if expressions, and loops! If you want to practice with the concepts discussed in this chapter, try building programs to do the following:

• Convert temperatures between Fahrenheit and Celsius.
• Generate the nth Fibonacci number.
• Print the lyrics to the Christmas carol “The Twelve Days of Christmas,” taking advantage of the repetition in the song.

When you’re ready to move on, we’ll talk about a concept in Rust that doesn’t commonly exist in other programming languages: ownership.

Ownership is Rust’s most unique feature, and it enables Rust to make memory safety guarantees without needing a garbage collector. Therefore, it’s important to understand how ownership works in Rust. In this chapter, we’ll talk about ownership as well as several related features: borrowing, slices, and how Rust lays data out in memory.

Rust’s central feature is ownership. Although the feature is straightforward to explain, it has deep implications for the rest of the language.

All programs have to manage the way they use a computer’s memory while running. Some languages have garbage collection that constantly looks for no longer used memory as the program runs; in other languages, the programmer must explicitly allocate and free the memory. Rust uses a third approach: memory is managed through a system of ownership with a set of rules that the compiler checks at compile time. None of the ownership features slow down your program while it's running.

Because ownership is a new concept for many programmers, it does take some time to get used to. The good news is that the more experienced you become with Rust and the rules of the ownership system, the more you’ll be able to naturally develop code that is safe and efficient. Keep at it!

When you understand ownership, you’ll have a solid foundation for understanding the features that make Rust unique. In this chapter, you’ll learn ownership by working through some examples that focus on a very common data structure: strings.

The Stack and the Heap

In many programming languages, you don’t have to think about the stack and the heap very often. But in a systems programming language like Rust, whether a value is on the stack or the heap has more of an effect on how the language behaves and why you have to make certain decisions. Parts of ownership will be described in relation to the stack and the heap later in this chapter, so here is a brief explanation in preparation.

Both the stack and the heap are parts of memory that is available to your code to use at runtime, but they are structured in different ways. The stack stores values in the order it gets them and removes the values in the opposite order. This is referred to as last in, first out. Think of a stack of plates: when you add more plates, you put them on top of the pile, and when you need a plate, you take one off the top. Adding or removing plates from the middle or bottom wouldn’t work as well! Adding data is called pushing onto the stack, and removing data is called popping off the stack.

The stack is fast because of the way it accesses the data: it never has to search for a place to put new data or a place to get data from because that place is always the top. Another property that makes the stack fast is that all data on the stack must take up a known, fixed size.

Data with a size unknown at compile time or a size that might change can be stored on the heap instead. The heap is less organized: when you put data on the heap, you ask for some amount of space. The operating system finds an empty spot somewhere in the heap that is big enough, marks it as being in use, and returns a pointer, which is the address of that location. This process is called allocating on the heap, sometimes abbreviated as just “allocating.” Pushing values onto the stack is not considered allocating. Because the pointer is a known, fixed size, you can store the pointer on the stack, but when you want the actual data, you have to follow the pointer.

Think of being seated at a restaurant. When you enter, you state the number of people in your group, and the staff finds an empty table that fits everyone and leads you there. If someone in your group comes late, they can ask where you’ve been seated to find you.

Accessing data in the heap is slower than accessing data on the stack because you have to follow a pointer to get there. Contemporary processors are faster if they jump around less in memory. Continuing the analogy, consider a server at a restaurant taking orders from many tables. It’s most efficient to get all the orders at one table before moving on to the next table. Taking an order from table A, then an order from table B, then one from A again, and then one from B again would be a much slower process. By the same token, a processor can do its job better if it works on data that’s close to other data (as it is on the stack) rather than farther away (as it can be on the heap). Allocating a large amount of space on the heap can also take time.

When your code calls a function, the values passed into the function (including, potentially, pointers to data on the heap) and the function’s local variables get pushed onto the stack. When the function is over, those values get popped off the stack.

Keeping track of what parts of code are using what data on the heap, minimizing the amount of duplicate data on the heap, and cleaning up unused data on the heap so you don’t run out of space are all problems that ownership addresses. Once you understand ownership, you won’t need to think about the stack and the heap very often, but knowing that managing heap data is why ownership exists can help explain why it works the way it does.

First, let’s take a look at the ownership rules. Keep these rules in mind as we work through the examples that illustrate them:

1. Each value in Rust has a variable that’s called its owner.
2. There can only be one owner at a time.
3. When the owner goes out of scope, the value will be dropped.

We’ve walked through an example of a Rust program already in Chapter 2. Now that we’re past basic syntax, we won’t include all the fn main() { code in examples, so if you’re following along, you’ll have to put the following examples inside a main function manually. As a result, our examples will be a bit more concise, letting us focus on the actual details rather than boilerplate code.

As a first example of ownership, we’ll look at the scope of some variables. A scope is the range within a program for which an item is valid. Let’s say we have a variable that looks like this:

# #![allow(unused_variables)]
#fn main() {
let s = "hello";
#}

The variable s refers to a string literal, where the value of the string is hardcoded into the text of our program. The variable is valid from the point at which it’s declared until the end of the current scope. Listing 4-1 has comments annotating where the variable s is valid:

# #![allow(unused_variables)]
#fn main() {
{                      // s is not valid here, it’s not yet declared
    let s = "hello";   // s is valid from this point forward

    // do stuff with s
}                      // this scope is now over, and s is no longer valid
#}

Listing 4-1: A variable and the scope in which it is valid

In other words, there are two important points in time here:

• When s comes into scope, it is valid.
• It remains valid until it goes out of scope.

At this point, the relationship between scopes and when variables are valid is similar to that in other programming languages. Now we’ll build on top of this understanding by introducing the String type.

To illustrate the rules of ownership, we need a data type that is more complex than the ones we covered in the “Data Types” section of Chapter 3. The types covered previously are all stored on the stack and popped off the stack when their scope is over, but we want to look at data that is stored on the heap and explore how Rust knows when to clean up that data.

We’ll use String as the example here and concentrate on the parts of String that relate to ownership. These aspects also apply to other complex data types provided by the standard library and that you create. We’ll discuss String in more depth in Chapter 8.

We’ve already seen string literals, where a string value is hardcoded into our program. String literals are convenient, but they aren’t suitable for every situation in which we may want to use text. One reason is that they’re immutable. Another is that not every string value can be known when we write our code: for example, what if we want to take user input and store it? For these situations, Rust has a second string type, String. This type is allocated on the heap and as such is able to store an amount of text that is unknown to us at compile time. You can create a String from a string literal using the from function, like so:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("hello");
#}

The double colon (::) is an operator that allows us to namespace this particular from function under the String type rather than using some sort of name like string_from. We’ll discuss this syntax more in the “Method Syntax” section of Chapter 5 and when we talk about namespacing with modules in “Module Definitions” in Chapter 7.

This kind of string can be mutated:

# #![allow(unused_variables)]
#fn main() {
let mut s = String::from("hello");

s.push_str(", world!"); // push_str() appends a literal to a String

println!("{}", s); // This will print `hello, world!`
#}

So, what’s the difference here? Why can String be mutated but literals cannot? The difference is how these two types deal with memory.

In the case of a string literal, we know the contents at compile time, so the text is hardcoded directly into the final executable. This is why string literals are fast and efficient. But these properties only come from the string literal’s immutability. Unfortunately, we can’t put a blob of memory into the binary for each piece of text whose size is unknown at compile time and whose size might change while running the program.

With the String type, in order to support a mutable, growable piece of text, we need to allocate an amount of memory on the heap, unknown at compile time, to hold the contents. This means:

• The memory must be requested from the operating system at runtime.
• We need a way of returning this memory to the operating system when we’re done with our String.

That first part is done by us: when we call String::from, its implementation requests the memory it needs. This is pretty much universal in programming languages.

However, the second part is different. In languages with a garbage collector (GC), the GC keeps track and cleans up memory that isn’t being used anymore, and we don’t need to think about it. Without a GC, it’s our responsibility to identify when memory is no longer being used and call code to explicitly return it, just as we did to request it. Doing this correctly has historically been a difficult programming problem. If we forget, we’ll waste memory. If we do it too early, we’ll have an invalid variable. If we do it twice, that’s a bug too. We need to pair exactly one allocate with exactly one free.

Rust takes a different path: the memory is automatically returned once the variable that owns it goes out of scope. Here’s a version of our scope example from Listing 4-1 using a String instead of a string literal:

# #![allow(unused_variables)]
#fn main() {
{
    let s = String::from("hello"); // s is valid from this point forward

    // do stuff with s
}                                  // this scope is now over, and s is no
                                   // longer valid
#}

There is a natural point at which we can return the memory our String needs to the operating system: when s goes out of scope. When a variable goes out of scope, Rust calls a special function for us. This function is called drop, and it’s where the author of String can put the code to return the memory. Rust calls drop automatically at the closing }.

Note: In C++, this pattern of deallocating resources at the end of an item’s lifetime is sometimes called Resource Acquisition Is Initialization (RAII). The drop function in Rust will be familiar to you if you’ve used RAII patterns.

This pattern has a profound impact on the way Rust code is written. It may seem simple right now, but the behavior of code can be unexpected in more complicated situations when we want to have multiple variables use the data we’ve allocated on the heap. Let’s explore some of those situations now.

Multiple variables can interact with the same data in different ways in Rust. Let’s look at an example using an integer in Listing 4-2:

# #![allow(unused_variables)]
#fn main() {
let x = 5;
let y = x;
#}

Listing 4-2: Assigning the integer value of variable x to y

We can probably guess what this is doing: “bind the value 5 to x; then make a copy of the value in x and bind it to y.” We now have two variables, x and y, and both equal 5. This is indeed what is happening, because integers are simple values with a known, fixed size, and these two 5 values are pushed onto the stack.

Now let’s look at the String version:

# #![allow(unused_variables)]
#fn main() {
let s1 = String::from("hello");
let s2 = s1;
#}

This looks very similar to the previous code, so we might assume that the way it works would be the same: that is, the second line would make a copy of the value in s1 and bind it to s2. But this isn’t quite what happens.

Take a look at Figure 4-1 to see what is happening to String under the covers. A String is made up of three parts, shown on the left: a pointer to the memory that holds the contents of the string, a length, and a capacity. This group of data is stored on the stack. On the right is the memory on the heap that holds the contents.

Figure 4-1: Representation in memory of a String holding the value "hello" bound to s1

The length is how much memory, in bytes, the contents of the String is currently using. The capacity is the total amount of memory, in bytes, that the String has received from the operating system. The difference between length and capacity matters, but not in this context, so for now, it’s fine to ignore the capacity.

When we assign s1 to s2, the String data is copied, meaning we copy the pointer, the length, and the capacity that are on the stack. We do not copy the data on the heap that the pointer refers to. In other words, the data representation in memory looks like Figure 4-2.

Figure 4-2: Representation in memory of the variable s2 that has a copy of the pointer, length, and capacity of s1

The representation does not look like Figure 4-3, which is what memory would look like if Rust instead copied the heap data as well. If Rust did this, the operation s2 = s1 could be very expensive in terms of runtime performance if the data on the heap were large.

Figure 4-3: Another possibility for what s2 = s1 might do if Rust copied the heap data as well

Earlier, we said that when a variable goes out of scope, Rust automatically calls the drop function and cleans up the heap memory for that variable. But Figure 4-2 shows both data pointers pointing to the same location. This is a problem: when s2 and s1 go out of scope, they will both try to free the same memory. This is known as a double free error and is one of the memory safety bugs we mentioned previously. Freeing memory twice can lead to memory corruption, which can potentially lead to security vulnerabilities.

To ensure memory safety, there’s one more detail to what happens in this situation in Rust. Instead of trying to copy the allocated memory, Rust considers s1 to no longer be valid and, therefore, Rust doesn’t need to free anything when s1 goes out of scope. Check out what happens when you try to use s1 after s2 is created; it won’t work:

let s1 = String::from("hello");
let s2 = s1;

println!("{}, world!", s1);

You’ll get an error like this because Rust prevents you from using the invalidated reference:

error[E0382]: use of moved value: `s1`
 --> src/main.rs:5:28
  |
3 |     let s2 = s1;
  |         -- value moved here
4 |
5 |     println!("{}, world!", s1);
  |                            ^^ value used here after move
  |
  = note: move occurs because `s1` has type `std::string::String`, which does
  not implement the `Copy` trait

If you’ve heard the terms shallow copy and deep copy while working with other languages, the concept of copying the pointer, length, and capacity without copying the data probably sounds like making a shallow copy. But because Rust also invalidates the first variable, instead of being called a shallow copy, it’s known as a move. Here we would read this by saying that s1 was moved into s2. So what actually happens is shown in Figure 4-4.

Figure 4-4: Representation in memory after s1 has been invalidated

That solves our problem! With only s2 valid, when it goes out of scope, it alone will free the memory, and we’re done.

In addition, there’s a design choice that’s implied by this: Rust will never automatically create “deep” copies of your data. Therefore, any automatic copying can be assumed to be inexpensive in terms of runtime performance.

If we do want to deeply copy the heap data of the String, not just the stack data, we can use a common method called clone. We’ll discuss method syntax in Chapter 5, but because methods are a common feature in many programming languages, you’ve probably seen them before.

Here’s an example of the clone method in action:

# #![allow(unused_variables)]
#fn main() {
let s1 = String::from("hello");
let s2 = s1.clone();

println!("s1 = {}, s2 = {}", s1, s2);
#}

This works just fine and explicitly produces the behavior shown in Figure 4-3, where the heap data does get copied.

When you see a call to clone, you know that some arbitrary code is being executed and that code may be expensive. It’s a visual indicator that something different is going on.

There’s another wrinkle we haven’t talked about yet. This code using integers, part of which was shown earlier in Listing 4-2, works and is valid:

# #![allow(unused_variables)]
#fn main() {
let x = 5;
let y = x;

println!("x = {}, y = {}", x, y);
#}

But this code seems to contradict what we just learned: we don’t have a call to clone, but x is still valid and wasn’t moved into y.

The reason is that types such as integers that have a known size at compile time are stored entirely on the stack, so copies of the actual values are quick to make. That means there’s no reason we would want to prevent x from being valid after we create the variable y. In other words, there’s no difference between deep and shallow copying here, so calling clone wouldn’t do anything different from the usual shallow copying and we can leave it out.

Rust has a special annotation called the Copy trait that we can place on types like integers that are stored on the stack (we’ll talk more about traits in Chapter 10). If a type has the Copy trait, an older variable is still usable after assignment. Rust won’t let us annotate a type with the Copy trait if the type, or any of its parts, has implemented the Drop trait. If the type needs something special to happen when the value goes out of scope and we add the Copy annotation to that type, we’ll get a compile time error. To learn about how to add the Copy annotation to your type, see “Derivable Traits” in Appendix C.

So what types are Copy? You can check the documentation for the given type to be sure, but as a general rule, any group of simple scalar values can be Copy, and nothing that requires allocation or is some form of resource is Copy. Here are some of the types that are Copy:

• All the integer types, such as u32.
• The Boolean type, bool, with values true and false.
• All the floating point types, such as f64.
• The character type, char.
• Tuples, but only if they contain types that are also Copy. For example, (i32, i32) is Copy, but (i32, String) is not.

The semantics for passing a value to a function are similar to those for assigning a value to a variable. Passing a variable to a function will move or copy, just as assignment does. Listing 4-3 has an example with some annotations showing where variables go into and out of scope:

Filename: src/main.rs

fn main() {
    let s = String::from("hello");  // s comes into scope

    takes_ownership(s);             // s's value moves into the function...
                                    // ... and so is no longer valid here

    let x = 5;                      // x comes into scope

    makes_copy(x);                  // x would move into the function,
                                    // but i32 is Copy, so it’s okay to still
                                    // use x afterward

} // Here, x goes out of scope, then s. But because s's value was moved, nothing
  // special happens.

fn takes_ownership(some_string: String) { // some_string comes into scope
    println!("{}", some_string);
} // Here, some_string goes out of scope and `drop` is called. The backing
  // memory is freed.

fn makes_copy(some_integer: i32) { // some_integer comes into scope
    println!("{}", some_integer);
} // Here, some_integer goes out of scope. Nothing special happens.

Listing 4-3: Functions with ownership and scope annotated

If we tried to use s after the call to takes_ownership, Rust would throw a compile time error. These static checks protect us from mistakes. Try adding code to main that uses s and x to see where you can use them and where the ownership rules prevent you from doing so.

Returning values can also transfer ownership. Listing 4-4 is an example with similar annotations to those in Listing 4-3:

Filename: src/main.rs

fn main() {
    let s1 = gives_ownership();         // gives_ownership moves its return
                                        // value into s1

    let s2 = String::from("hello");     // s2 comes into scope

    let s3 = takes_and_gives_back(s2);  // s2 is moved into
                                        // takes_and_gives_back, which also
                                        // moves its return value into s3
} // Here, s3 goes out of scope and is dropped. s2 goes out of scope but was
  // moved, so nothing happens. s1 goes out of scope and is dropped.

fn gives_ownership() -> String {             // gives_ownership will move its
                                             // return value into the function
                                             // that calls it

    let some_string = String::from("hello"); // some_string comes into scope

    some_string                              // some_string is returned and
                                             // moves out to the calling
                                             // function.
}

// takes_and_gives_back will take a String and return one.
fn takes_and_gives_back(a_string: String) -> String { // a_string comes into
                                                      // scope

    a_string  // a_string is returned and moves out to the calling function
}

Listing 4-4: Transferring ownership of return values

The ownership of a variable follows the same pattern every time: assigning a value to another variable moves it. When a variable that includes data on the heap goes out of scope, the value will be cleaned up by drop unless the data has been moved to be owned by another variable.

Taking ownership and then returning ownership with every function is a bit tedious. What if we want to let a function use a value but not take ownership? It’s quite annoying that anything we pass in also needs to be passed back if we want to use it again, in addition to any data resulting from the body of the function that we might want to return as well.

It’s possible to return multiple values using a tuple, as shown in Listing 4-5:

Filename: src/main.rs

fn main() {
    let s1 = String::from("hello");

    let (s2, len) = calculate_length(s1);

    println!("The length of '{}' is {}.", s2, len);
}

fn calculate_length(s: String) -> (String, usize) {
    let length = s.len(); // len() returns the length of a String

    (s, length)
}

Listing 4-5: Returning ownership of parameters

But this is too much ceremony and a lot of work for a concept that should be common. Luckily for us, Rust has a feature for this concept, called references.

The issue with the tuple code in Listing 4-5 is that we have to return the String to the calling function so we can still use the String after the call to calculate_length, because the String was moved into calculate_length.

Here is how you would define and use a calculate_length function that has a reference to an object as a parameter instead of taking ownership of the value:

Filename: src/main.rs

fn main() {
    let s1 = String::from("hello");

    let len = calculate_length(&s1);

    println!("The length of '{}' is {}.", s1, len);
}

fn calculate_length(s: &String) -> usize {
    s.len()
}

First, notice that all the tuple code in the variable declaration and the function return value is gone. Second, note that we pass &s1 into calculate_length and, in its definition, we take &String rather than String.

These ampersands are references, and they allow you to refer to some value without taking ownership of it. Figure 4-5 shows a diagram.

Figure 4-5: A diagram of &String s pointing at String s1

Note: The opposite of referencing by using & is dereferencing, which is accomplished with the dereference operator, *. We’ll see some uses of the dereference operator in Chapter 8 and discuss details of dereferencing in Chapter 15.

Let’s take a closer look at the function call here:

# #![allow(unused_variables)]
#fn main() {
# fn calculate_length(s: &String) -> usize {
#     s.len()
# }
let s1 = String::from("hello");

let len = calculate_length(&s1);
#}

The &s1 syntax lets us create a reference that refers to the value of s1 but does not own it. Because it does not own it, the value it points to will not be dropped when the reference goes out of scope.

Likewise, the signature of the function uses & to indicate that the type of the parameter s is a reference. Let’s add some explanatory annotations:

# #![allow(unused_variables)]
#fn main() {
fn calculate_length(s: &String) -> usize { // s is a reference to a String
    s.len()
} // Here, s goes out of scope. But because it does not have ownership of what
  // it refers to, nothing happens.
#}

The scope in which the variable s is valid is the same as any function parameter’s scope, but we don’t drop what the reference points to when it goes out of scope because we don’t have ownership. When functions have references as parameters instead of the actual values, we won’t need to return the values in order to give back ownership, because we never had ownership.

We call having references as function parameters borrowing. As in real life, if a person owns something, you can borrow it from them. When you’re done, you have to give it back.

So what happens if we try to modify something we’re borrowing? Try the code in Listing 4-6. Spoiler alert: it doesn’t work!

Filename: src/main.rs

fn main() {
    let s = String::from("hello");

    change(&s);
}

fn change(some_string: &String) {
    some_string.push_str(", world");
}

Listing 4-6: Attempting to modify a borrowed value

Here’s the error:

error[E0596]: cannot borrow immutable borrowed content `*some_string` as mutable
 --> error.rs:8:5
  |
7 | fn change(some_string: &String) {
  |                        ------- use `&mut String` here to make mutable
8 |     some_string.push_str(", world");
  |     ^^^^^^^^^^^ cannot borrow as mutable

Just as variables are immutable by default, so are references. We’re not allowed to modify something we have a reference to.

We can fix the error in the code from Listing 4-6 with just a small tweak:

Filename: src/main.rs

fn main() {
    let mut s = String::from("hello");

    change(&mut s);
}

fn change(some_string: &mut String) {
    some_string.push_str(", world");
}

First, we had to change s to be mut. Then we had to create a mutable reference with &mut s and accept a mutable reference with some_string: &mut String.

But mutable references have one big restriction: you can only have one mutable reference to a particular piece of data in a particular scope. This code will fail:

Filename: src/main.rs

let mut s = String::from("hello");

let r1 = &mut s;
let r2 = &mut s;

Here’s the error:

error[E0499]: cannot borrow `s` as mutable more than once at a time
 --> borrow_twice.rs:5:19
  |
4 |     let r1 = &mut s;
  |                   - first mutable borrow occurs here
5 |     let r2 = &mut s;
  |                   ^ second mutable borrow occurs here
6 | }
  | - first borrow ends here

This restriction allows for mutation but in a very controlled fashion. It’s something that new Rustaceans struggle with, because most languages let you mutate whenever you’d like.

The benefit of having this restriction is that Rust can prevent data races at compile time. A data race is similar to a race condition and happens when these three behaviors occur:

• Two or more pointers access the same data at the same time.
• At least one of the pointers is being used to write to the data.
• There’s no mechanism being used to synchronize access to the data.

Data races cause undefined behavior and can be difficult to diagnose and fix when you’re trying to track them down at runtime; Rust prevents this problem from happening because it won’t even compile code with data races!

As always, we can use curly brackets to create a new scope, allowing for multiple mutable references, just not simultaneous ones:

# #![allow(unused_variables)]
#fn main() {
let mut s = String::from("hello");

{
    let r1 = &mut s;

} // r1 goes out of scope here, so we can make a new reference with no problems.

let r2 = &mut s;
#}

A similar rule exists for combining mutable and immutable references. This code results in an error:

let mut s = String::from("hello");

let r1 = &s; // no problem
let r2 = &s; // no problem
let r3 = &mut s; // BIG PROBLEM

Here’s the error:

error[E0502]: cannot borrow `s` as mutable because it is also borrowed as
immutable
 --> borrow_thrice.rs:6:19
  |
4 |     let r1 = &s; // no problem
  |               - immutable borrow occurs here
5 |     let r2 = &s; // no problem
6 |     let r3 = &mut s; // BIG PROBLEM
  |                   ^ mutable borrow occurs here
7 | }
  | - immutable borrow ends here

Whew! We also cannot have a mutable reference while we have an immutable one. Users of an immutable reference don’t expect the values to suddenly change out from under them! However, multiple immutable references are okay because no one who is just reading the data has the ability to affect anyone else’s reading of the data.

Even though these errors may be frustrating at times, remember that it’s the Rust compiler pointing out a potential bug early (at compile time rather than at runtime) and showing you exactly where the problem is. Then you don’t have to track down why your data isn’t what you thought it was.

In languages with pointers, it’s easy to erroneously create a dangling pointer, a pointer that references a location in memory that may have been given to someone else, by freeing some memory while preserving a pointer to that memory. In Rust, by contrast, the compiler guarantees that references will never be dangling references: if you have a reference to some data, the compiler will ensure that the data will not go out of scope before the reference to the data does.

Let’s try to create a dangling reference, which Rust will prevent with a compile-time error:

Filename: src/main.rs

fn main() {
    let reference_to_nothing = dangle();
}

fn dangle() -> &String {
    let s = String::from("hello");

    &s
}

Here’s the error:

error[E0106]: missing lifetime specifier
 --> dangle.rs:5:16
  |
5 | fn dangle() -> &String {
  |                ^ expected lifetime parameter
  |
  = help: this function's return type contains a borrowed value, but there is
  no value for it to be borrowed from
  = help: consider giving it a 'static lifetime

This error message refers to a feature we haven’t covered yet: lifetimes. We’ll discuss lifetimes in detail in Chapter 10. But, if you disregard the parts about lifetimes, the message does contain the key to why this code is a problem:

this function's return type contains a borrowed value, but there is no value
for it to be borrowed from.

Let’s take a closer look at exactly what’s happening at each stage of our dangle code:

fn dangle() -> &String { // dangle returns a reference to a String

    let s = String::from("hello"); // s is a new String

    &s // we return a reference to the String, s
} // Here, s goes out of scope, and is dropped. Its memory goes away.
  // Danger!

Because s is created inside dangle, when the code of dangle is finished, s will be deallocated. But we tried to return a reference to it. That means this reference would be pointing to an invalid String That’s no good! Rust won’t let us do this.

The solution here is to return the String directly:

# #![allow(unused_variables)]
#fn main() {
fn no_dangle() -> String {
    let s = String::from("hello");

    s
}
#}

This works without any problems. Ownership is moved out, and nothing is deallocated.

Let’s recap what we’ve discussed about references:

• At any given time, you can have either (but not both of) one mutable reference or any number of immutable references.
• References must always be valid.

Next, we’ll look at a different kind of reference: slices.

Another data type that does not have ownership is the slice. Slices let you reference a contiguous sequence of elements in a collection rather than the whole collection.

Here’s a small programming problem: write a function that takes a string and returns the first word it finds in that string. If the function doesn’t find a space in the string, the whole string must be one word, so the entire string should be returned.

Let’s think about the signature of this function:

fn first_word(s: &String) -> ?

This function, first_word, has a &String as a parameter. We don’t want ownership, so this is fine. But what should we return? We don’t really have a way to talk about part of a string. However, we could return the index of the end of the word. Let’s try that, as shown in Listing 4-7:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
fn first_word(s: &String) -> usize {
    let bytes = s.as_bytes();

    for (i, &item) in bytes.iter().enumerate() {
        if item == b' ' {
            return i;
        }
    }

    s.len()
}
#}

Listing 4-7: The first_word function that returns a byte index value into the String parameter

Because we need to go through the String element by element and check whether a value is a space, we’ll convert our String to an array of bytes using the as_bytes method:

let bytes = s.as_bytes();

Next, we create an iterator over the array of bytes using the iter method:

for (i, &item) in bytes.iter().enumerate() {

We’ll discuss iterators in more detail in Chapter 13. For now, know that iter is a method that returns each element in a collection and that enumerate wraps the result of iter and returns each element as part of a tuple instead. The first element of the tuple returned from enumerate is the index, and the second element is a reference to the element. This is a bit more convenient than calculating the index ourselves.

Because the enumerate method returns a tuple, we can use patterns to destructure that tuple, just like everywhere else in Rust. So in the for loop, we specify a pattern that has i for the index in the tuple and &item for the single byte in the tuple. Because we get a reference to the element from .iter().enumerate(), we use & in the pattern.

Inside the for loop, we search for the byte that represents the space by using the byte literal syntax. If we find a space, we return the position. Otherwise, we return the length of the string by using s.len():

    if item == b' ' {
        return i;
    }
}
s.len()

We now have a way to find out the index of the end of the first word in the string, but there’s a problem. We’re returning a usize on its own, but it’s only a meaningful number in the context of the &String. In other words, because it’s a separate value from the String, there’s no guarantee that it will still be valid in the future. Consider the program in Listing 4-8 that uses the first_word function from Listing 4-7:

Filename: src/main.rs

# fn first_word(s: &String) -> usize {
#     let bytes = s.as_bytes();
#
#     for (i, &item) in bytes.iter().enumerate() {
#         if item == b' ' {
#             return i;
#         }
#     }
#
#     s.len()
# }
#
fn main() {
    let mut s = String::from("hello world");

    let word = first_word(&s); // word will get the value 5

    s.clear(); // This empties the String, making it equal to ""

    // word still has the value 5 here, but there's no more string that
    // we could meaningfully use the value 5 with. word is now totally invalid!
}

Listing 4-8: Storing the result from calling the first_word function and then changing the String contents

This program compiles without any errors and would also do so if we used word after calling s.clear(). Because word isn’t connected to the state of s at all, word still contains the value 5. We could use that value 5 with the variable s to try to extract the first word out, but this would be a bug because the contents of s have changed since we saved 5 in word.

Having to worry about the index in word getting out of sync with the data in s is tedious and error prone! Managing these indices is even more brittle if we write a second_word function. Its signature would have to look like this:

fn second_word(s: &String) -> (usize, usize) {

Now we’re tracking a starting and an ending index, and we have even more values that were calculated from data in a particular state but aren’t tied to that state at all. We now have three unrelated variables floating around that need to be kept in sync.

Luckily, Rust has a solution to this problem: string slices.

A string slice is a reference to part of a String, and it looks like this:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("hello world");

let hello = &s[0..5];
let world = &s[6..11];
#}

This is similar to taking a reference to the whole String but with the extra [0..5] bit. Rather than a reference to the entire String, it’s a reference to a portion of the String. The start..end syntax is a range that begins at start and continues up to, but not including, end.

We can create slices using a range within brackets by specifying [starting_index..ending_index], where starting_index is the first position in the slice and ending_index is one more than the last position in the slice. Internally, the slice data structure stores the starting position and the length of the slice, which corresponds to ending_index minus starting_index. So in the case of let world = &s[6..11];, world would be a slice that contains a pointer to the 6th byte of s and a length value of 5.

Figure 4-6 shows this in a diagram.

Figure 4-6: String slice referring to part of a String

With Rust’s .. range syntax, if you want to start at the first index (zero), you can drop the value before the two periods. In other words, these are equal:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("hello");

let slice = &s[0..2];
let slice = &s[..2];
#}

By the same token, if your slice includes the last byte of the String, you can drop the trailing number. That means these are equal:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("hello");

let len = s.len();

let slice = &s[3..len];
let slice = &s[3..];
#}

You can also drop both values to take a slice of the entire string. So these are equal:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("hello");

let len = s.len();

let slice = &s[0..len];
let slice = &s[..];
#}

Note: String slice range indices must occur at valid UTF-8 character boundaries. If you attempt to create a string slice in the middle of a multibyte character, your program will exit with an error. For the purposes of introducing string slices, we are assuming ASCII only in this section; a more thorough discussion of UTF-8 handling is in the “Strings” section of Chapter 8.

With all this information in mind, let’s rewrite first_word to return a slice. The type that signifies “string slice” is written as &str:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
fn first_word(s: &String) -> &str {
    let bytes = s.as_bytes();

    for (i, &item) in bytes.iter().enumerate() {
        if item == b' ' {
            return &s[0..i];
        }
    }

    &s[..]
}
#}

We get the index for the end of the word in the same way as we did in Listing 4-7, by looking for the first occurrence of a space. When we find a space, we return a string slice using the start of the string and the index of the space as the starting and ending indices.

Now when we call first_word, we get back a single value that is tied to the underlying data. The value is made up of a reference to the starting point of the slice and the number of elements in the slice.

Returning a slice would also work for a second_word function:

fn second_word(s: &String) -> &str {

We now have a straightforward API that’s much harder to mess up, because the compiler will ensure the references into the String remain valid. Remember the bug in the program in Listing 4-8, when we got the index to the end of the first word but then cleared the string so our index was invalid? That code was logically incorrect but didn’t show any immediate errors. The problems would show up later if we kept trying to use the first word index with an emptied string. Slices make this bug impossible and let us know we have a problem with our code much sooner. Using the slice version of first_word will throw a compile time error:

Filename: src/main.rs

fn main() {
    let mut s = String::from("hello world");

    let word = first_word(&s);

    s.clear(); // Error!
}

Here’s the compiler error:

error[E0502]: cannot borrow `s` as mutable because it is also borrowed as immutable
 --> src/main.rs:6:5
  |
4 |     let word = first_word(&s);
  |                            - immutable borrow occurs here
5 |
6 |     s.clear(); // Error!
  |     ^ mutable borrow occurs here
7 | }
  | - immutable borrow ends here

Recall from the borrowing rules that if we have an immutable reference to something, we cannot also take a mutable reference. Because clear needs to truncate the String, it tries to take a mutable reference, which fails. Not only has Rust made our API easier to use, but it has also eliminated an entire class of errors at compile time!

Recall that we talked about string literals being stored inside the binary. Now that we know about slices, we can properly understand string literals:

# #![allow(unused_variables)]
#fn main() {
let s = "Hello, world!";
#}

The type of s here is &str: it’s a slice pointing to that specific point of the binary. This is also why string literals are immutable; &str is an immutable reference.

Knowing that you can take slices of literals and Strings leads us to one more improvement on first_word, and that’s its signature:

fn first_word(s: &String) -> &str {

A more experienced Rustacean would write the following line instead because it allows us to use the same function on both Strings and &strs:

fn first_word(s: &str) -> &str {

If we have a string slice, we can pass that directly. If we have a String, we can pass a slice of the entire String. Defining a function to take a string slice instead of a reference to a String makes our API more general and useful without losing any functionality:

Filename: src/main.rs

# fn first_word(s: &str) -> &str {
#     let bytes = s.as_bytes();
#
#     for (i, &item) in bytes.iter().enumerate() {
#         if item == b' ' {
#             return &s[0..i];
#         }
#     }
#
#     &s[..]
# }
fn main() {
    let my_string = String::from("hello world");

    // first_word works on slices of `String`s
    let word = first_word(&my_string[..]);

    let my_string_literal = "hello world";

    // first_word works on slices of string literals
    let word = first_word(&my_string_literal[..]);

    // Because string literals *are* string slices already,
    // this works too, without the slice syntax!
    let word = first_word(my_string_literal);
}

String slices, as you might imagine, are specific to strings. But there’s a more general slice type, too. Consider this array:

# #![allow(unused_variables)]
#fn main() {
let a = [1, 2, 3, 4, 5];
#}

Just as we might want to refer to a part of a string, we might want to refer to part of an array. We’d do so like this:

# #![allow(unused_variables)]
#fn main() {
let a = [1, 2, 3, 4, 5];

let slice = &a[1..3];
#}

This slice has the type &[i32]. It works the same way as string slices do, by storing a reference to the first element and a length. You’ll use this kind of slice for all sorts of other collections. We’ll discuss these collections in detail when we talk about vectors in Chapter 8.

The concepts of ownership, borrowing, and slices ensure memory safety in Rust programs at compile time. The Rust language gives you control over your memory usage in the same way as other systems programming languages, but having the owner of data automatically clean up that data when the owner goes out of scope means you don’t have to write and debug extra code to get this control.

Ownership affects how lots of other parts of Rust work, so we’ll talk about these concepts further throughout the rest of the book. Let’s move on to Chapter 5 and look at grouping pieces of data together in a struct.

A struct, or structure, is a custom data type that lets you name and package together multiple related values that make up a meaningful group. If you’re familiar with an object-oriented language, a struct is like an object’s data attributes. In this chapter, we’ll compare and contrast tuples with structs, demonstrate how to use structs, and discuss how to define methods and associated functions to specify behavior associated with a struct’s data. Structs and enums (discussed in Chapter 6) are the building blocks for creating new types in your program’s domain to take full advantage of Rust’s compile time type checking.

Structs are similar to tuples, which were discussed in Chapter 3. Like tuples, the pieces of a struct can be different types. Unlike with tuples, you’ll name each piece of data so it’s clear what the values mean. As a result of these names, structs are more flexible than tuples: you don’t have to rely on the order of the data to specify or access the values of an instance.

To define a struct, we enter the keyword struct and name the entire struct. A struct’s name should describe the significance of the pieces of data being grouped together. Then, inside curly brackets, we define the names and types of the pieces of data, which we call fields. For example, Listing 5-1 shows a struct that stores information about a user account:

# #![allow(unused_variables)]
#fn main() {
struct User {
    username: String,
    email: String,
    sign_in_count: u64,
    active: bool,
}
#}

Listing 5-1: A User struct definition

To use a struct after we’ve defined it, we create an instance of that struct by specifying concrete values for each of the fields. We create an instance by stating the name of the struct and then add curly brackets containing key: value pairs, where the keys are the names of the fields and the values are the data we want to store in those fields. We don’t have to specify the fields in the same order in which we declared them in the struct. In other words, the struct definition is like a general template for the type, and instances fill in that template with particular data to create values of the type. For example, we can declare a particular user as shown in Listing 5-2:

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
let user1 = User {
    email: String::from("someone@example.com"),
    username: String::from("someusername123"),
    active: true,
    sign_in_count: 1,
};
#}

Listing 5-2: Creating an instance of the User struct

To get a specific value from a struct, we can use dot notation. If we wanted just this user’s email address, we could use user1.email wherever we wanted to use this value. If the instance is mutable, we can change a value by using the dot notation and assigning into a particular field. Listing 5-3 shows how to change the value in the email field of a mutable User instance:

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
let mut user1 = User {
    email: String::from("someone@example.com"),
    username: String::from("someusername123"),
    active: true,
    sign_in_count: 1,
};

user1.email = String::from("anotheremail@example.com");
#}

Listing 5-3: Changing the value in the email field of a User instance

Note that the entire instance must be mutable; Rust doesn’t allow us to mark only certain fields as mutable.

As with any expression, we can construct a new instance of the struct as the last expression in the function body to implicitly return that new instance. Listing 5-4 shows a build_user function that returns a User instance with the given email and username. The active field gets the value of true, and the sign_in_count gets a value of 1.

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
fn build_user(email: String, username: String) -> User {
    User {
        email: email,
        username: username,
        active: true,
        sign_in_count: 1,
    }
}
#}

Listing 5-4: A build_user function that takes an email and username and returns a User instance

It makes sense to name the function parameters with the same name as the struct fields, but having to repeat the email and username field names and variables is a bit tedious. If the struct had more fields, repeating each name would get even more annoying. Luckily, there’s a convenient shorthand!

Because the parameter names and the struct field names are exactly the same in Listing 5-4, we can use the field init shorthand syntax to rewrite build_user so that it behaves exactly the same but doesn’t have the repetition of email and username as shown in Listing 5-5.

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
fn build_user(email: String, username: String) -> User {
    User {
        email,
        username,
        active: true,
        sign_in_count: 1,
    }
}
#}

Listing 5-5: A build_user function that uses field init shorthand because the email and username parameters have the same name as struct fields

Here, we’re creating a new instance of the User struct, which has a field named email. We want to set the email field’s value to the value in the email parameter of the build_user function. Because the email field and the email parameter have the same name, we only need to write email rather than email: email.

It’s often useful to create a new instance of a struct that uses most of an old instance’s values but changes some. You’ll do this using struct update syntax.

First, Listing 5-6 shows how we create a new User instance in user2 without the update syntax. We set new values for email and username but otherwise use the same values from user1 that we created in Listing 5-2:

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
# let user1 = User {
#     email: String::from("someone@example.com"),
#     username: String::from("someusername123"),
#     active: true,
#     sign_in_count: 1,
# };
#
let user2 = User {
    email: String::from("another@example.com"),
    username: String::from("anotherusername567"),
    active: user1.active,
    sign_in_count: user1.sign_in_count,
};
#}

Listing 5-6: Creating a new User instance using some of the values from user1

Using struct update syntax, we can achieve the same effect with less code, as shown in Listing 5-7. The syntax .. specifies that the remaining fields not explicitly set should have the same value as the fields in the given instance.

# #![allow(unused_variables)]
#fn main() {
# struct User {
#     username: String,
#     email: String,
#     sign_in_count: u64,
#     active: bool,
# }
#
# let user1 = User {
#     email: String::from("someone@example.com"),
#     username: String::from("someusername123"),
#     active: true,
#     sign_in_count: 1,
# };
#
let user2 = User {
    email: String::from("another@example.com"),
    username: String::from("anotherusername567"),
    ..user1
};
#}

Listing 5-7: Using struct update syntax to set new email and username values for a User instance but use the rest of the values from the fields of the instance in the user1 variable

The code in Listing 5-7 also creates an instance in user2 that has a different value for email and username but has the same values for the active and sign_in_count fields from user1.

You can also define structs that look similar to tuples, called tuple structs. Tuple structs have the added meaning the struct name provides but don’t have names associated with their fields; rather, they just have the types of the fields. Tuple structs are useful when you want to give the whole tuple a name and make the tuple be a different type than other tuples, and naming each field as in a regular struct would be verbose or redundant.

To define a tuple struct start with the struct keyword and the struct name followed by the types in the tuple. For example, here are definitions and usages of two tuple structs named Color and Point:

# #![allow(unused_variables)]
#fn main() {
struct Color(i32, i32, i32);
struct Point(i32, i32, i32);

let black = Color(0, 0, 0);
let origin = Point(0, 0, 0);
#}

Note that the black and origin values are different types, because they’re instances of different tuple structs. Each struct you define is its own type, even though the fields within the struct have the same types. For example, a function that takes a parameter of type Color cannot take a Point as an argument, even though both types are made up of three i32 values. Otherwise, tuple struct instances behave like tuples: you can destructure them into their individual pieces, you can use a . followed by the index to access an individual value, and so on.

You can also define structs that don’t have any fields! These are called unit-like structs because they behave similarly to (), the unit type. Unit-like structs can be useful in situations in which you need to implement a trait on some type but don’t have any data that you want to store in the type itself. We’ll discuss traits in Chapter 10.

Ownership of Struct Data

In the User struct definition in Listing 5-1, we used the owned String type rather than the &str string slice type. This is a deliberate choice because we want instances of this struct to own all of its data and for that data to be valid for as long as the entire struct is valid.

It’s possible for structs to store references to data owned by something else, but to do so requires the use of lifetimes, a Rust feature that we’ll discuss in Chapter 10. Lifetimes ensure that the data referenced by a struct is valid for as long as the struct is. Let’s say you try to store a reference in a struct without specifying lifetimes, like this, which won’t work:

Filename: src/main.rs

struct User {
    username: &str,
    email: &str,
    sign_in_count: u64,
    active: bool,
}

fn main() {
    let user1 = User {
        email: "someone@example.com",
        username: "someusername123",
        active: true,
        sign_in_count: 1,
    };
}

The compiler will complain that it needs lifetime specifiers:

error[E0106]: missing lifetime specifier
 -->
  |
2 |     username: &str,
  |               ^ expected lifetime parameter

error[E0106]: missing lifetime specifier
 -->
  |
3 |     email: &str,
  |            ^ expected lifetime parameter

In Chapter 10, we’ll discuss how to fix these errors so you can store references in structs, but for now, we’ll fix errors like these using owned types like String instead of references like &str.

To understand when we might want to use structs, let’s write a program that calculates the area of a rectangle. We’ll start with single variables, and then refactor the program until we’re using structs instead.

Let’s make a new binary project with Cargo called rectangles that will take the width and height of a rectangle specified in pixels and calculate the area of the rectangle. Listing 5-8 shows a short program with one way of doing exactly that in our project’s src/main.rs:

Filename: src/main.rs

fn main() {
    let width1 = 30;
    let height1 = 50;

    println!(
        "The area of the rectangle is {} square pixels.",
        area(width1, height1)
    );
}

fn area(width: u32, height: u32) -> u32 {
    width * height
}

Listing 5-8: Calculating the area of a rectangle specified by separate width and height variables

Now, run this program using cargo run:

The area of the rectangle is 1500 square pixels.

Even though Listing 5-8 works and figures out the area of the rectangle by calling the area function with each dimension, we can do better. The width and the height are related to each other because together they describe one rectangle.

The issue with this code is evident in the signature of area:

fn area(width: u32, height: u32) -> u32 {

The area function is supposed to calculate the area of one rectangle, but the function we wrote has two parameters. The parameters are related, but that’s not expressed anywhere in our program. It would be more readable and more manageable to group width and height together. We’ve already discussed one way we might do that in “The Tuple Type” section of Chapter 3: by using tuples.

Listing 5-9 shows another version of our program that uses tuples:

Filename: src/main.rs

fn main() {
    let rect1 = (30, 50);

    println!(
        "The area of the rectangle is {} square pixels.",
        area(rect1)
    );
}

fn area(dimensions: (u32, u32)) -> u32 {
    dimensions.0 * dimensions.1
}

Listing 5-9: Specifying the width and height of the rectangle with a tuple

In one way, this program is better. Tuples let us add a bit of structure, and we’re now passing just one argument. But in another way, this version is less clear: tuples don’t name their elements, so our calculation has become more confusing because we have to index into the parts of the tuple.

It doesn’t matter if we mix up width and height for the area calculation, but if we want to draw the rectangle on the screen, it would matter! We would have to keep in mind that width is the tuple index 0 and height is the tuple index 1. If someone else worked on this code, they would have to figure this out and keep it in mind as well. It would be easy to forget or mix up these values and cause errors, because we haven’t conveyed the meaning of our data in our code.

We use structs to add meaning by labeling the data. We can transform the tuple we’re using into a data type with a name for the whole as well as names for the parts, as shown in Listing 5-10:

Filename: src/main.rs

struct Rectangle {
    width: u32,
    height: u32,
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50 };

    println!(
        "The area of the rectangle is {} square pixels.",
        area(&rect1)
    );
}

fn area(rectangle: &Rectangle) -> u32 {
    rectangle.width * rectangle.height
}

Listing 5-10: Defining a Rectangle struct

Here we’ve defined a struct and named it Rectangle. Inside the curly brackets, we defined the fields as width and height, both of which have type u32. Then in main, we created a particular instance of Rectangle that has a width of 30 and a height of 50.

Our area function is now defined with one parameter, which we’ve named rectangle, whose type is an immutable borrow of a struct Rectangle instance. As mentioned in Chapter 4, we want to borrow the struct rather than take ownership of it. This way, main retains its ownership and can continue using rect1, which is the reason we use the & in the function signature and where we call the function.

The area function accesses the width and height fields of the Rectangle instance. Our function signature for area now says exactly what we mean: calculate the area of Rectangle, using its width and height fields. This conveys that the width and height are related to each other, and it gives descriptive names to the values rather than using the tuple index values of 0 and 1. This is a win for clarity.

It’d be nice to be able to print an instance of Rectangle while we’re debugging our program and see the values for all its fields. Listing 5-11 tries using the println! macro as we have used in previous chapters. This won’t work, however:

Filename: src/main.rs

struct Rectangle {
    width: u32,
    height: u32,
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50 };

    println!("rect1 is {}", rect1);
}

Listing 5-11: Attempting to print a Rectangle instance

When we run this code, we get an error with this core message:

error[E0277]: the trait bound `Rectangle: std::fmt::Display` is not satisfied

The println! macro can do many kinds of formatting, and by default, curly brackets tell println! to use formatting known as Display: output intended for direct end user consumption. The primitive types we’ve seen so far implement Display by default, because there’s only one way you’d want to show a 1 or any other primitive type to a user. But with structs, the way println! should format the output is less clear because there are more display possibilities: Do you want commas or not? Do you want to print the curly brackets? Should all the fields be shown? Due to this ambiguity, Rust doesn’t try to guess what we want, and structs don’t have a provided implementation of Display.

If we continue reading the errors, we’ll find this helpful note:

`Rectangle` cannot be formatted with the default formatter; try using
`:?` instead if you are using a format string

Let’s try it! The println! macro call will now look like println!("rect1 is {:?}", rect1);. Putting the specifier :? inside the curly brackets tells println! we want to use an output format called Debug. Debug is a trait that enables us to print our struct in a way that is useful for developers so we can see its value while we’re debugging our code.

Run the code with this change. Drat! We still get an error:

error[E0277]: the trait bound `Rectangle: std::fmt::Debug` is not satisfied

But again, the compiler gives us a helpful note:

`Rectangle` cannot be formatted using `:?`; if it is defined in your
crate, add `#[derive(Debug)]` or manually implement it

Rust does include functionality to print out debugging information, but we have to explicitly opt in to make that functionality available for our struct. To do that, we add the annotation #[derive(Debug)] just before the struct definition, as shown in Listing 5-12:

Filename: src/main.rs

#[derive(Debug)]
struct Rectangle {
    width: u32,
    height: u32,
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50 };

    println!("rect1 is {:?}", rect1);
}

Listing 5-12: Adding the annotation to derive the Debug trait and printing the Rectangle instance using debug formatting

Now when we run the program, we won’t get any errors, and we’ll see the following output:

rect1 is Rectangle { width: 30, height: 50 }

Nice! It’s not the prettiest output, but it shows the values of all the fields for this instance, which would definitely help during debugging. When we have larger structs, it’s useful to have output that’s a bit easier to read; in those cases, we can use {:#?} instead of {:?} in the println! string. When we use the {:#?} style in the example, the output will look like this:

rect1 is Rectangle {
    width: 30,
    height: 50
}

Rust has provided a number of traits for us to use with the derive annotation that can add useful behavior to our custom types. Those traits and their behaviors are listed in Appendix C, “Derivable Traits.” We’ll cover how to implement these traits with custom behavior as well as how to create your own traits in Chapter 10.

Our area function is very specific: it only computes the area of rectangles. It would be helpful to tie this behavior more closely to our Rectangle struct, because it won’t work with any other type. Let’s look at how we can continue to refactor this code by turning the area function into an area method defined on our Rectangle type.

Methods are similar to functions: they’re declared with the fn keyword and their name, they can have parameters and a return value, and they contain some code that is run when they’re called from somewhere else. However, methods are different from functions in that they’re defined within the context of a struct (or an enum or a trait object, which we cover in Chapters 6 and 17, respectively), and their first parameter is always self, which represents the instance of the struct the method is being called on.

Let’s change the area function that has a Rectangle instance as a parameter and instead make an area method defined on the Rectangle struct, as shown in Listing 5-13:

Filename: src/main.rs

#[derive(Debug)]
struct Rectangle {
    width: u32,
    height: u32,
}

impl Rectangle {
    fn area(&self) -> u32 {
        self.width * self.height
    }
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50 };

    println!(
        "The area of the rectangle is {} square pixels.",
        rect1.area()
    );
}

Listing 5-13: Defining an area method on the Rectangle struct

To define the function within the context of Rectangle, we start an impl (implementation) block. Then we move the area function within the impl curly brackets and change the first (and in this case, only) parameter to be self in the signature and everywhere within the body. In main, where we called the area function and passed rect1 as an argument, we can instead use method syntax to call the area method on our Rectangle instance. The method syntax goes after an instance: we add a dot followed by the method name, parentheses, and any arguments.

In the signature for area, we use &self instead of rectangle: &Rectangle because Rust knows the type of self is Rectangle due to this method’s being inside the impl Rectangle context. Note that we still need to use the & before self, just as we did in &Rectangle. Methods can take ownership of self, borrow self immutably as we’ve done here, or borrow self mutably, just as they can any other parameter.

We’ve chosen &self here for the same reason we used &Rectangle in the function version: we don’t want to take ownership, and we just want to read the data in the struct, not write to it. If we wanted to change the instance that we’ve called the method on as part of what the method does, we’d use &mut self as the first parameter. Having a method that takes ownership of the instance by using just self as the first parameter is rare; this technique is usually used when the method transforms self into something else and you want to prevent the caller from using the original instance after the transformation.

The main benefit of using methods instead of functions, in addition to using method syntax and not having to repeat the type of self in every method’s signature, is for organization. We’ve put all the things we can do with an instance of a type in one impl block rather than making future users of our code search for capabilities of Rectangle in various places in the library we provide.

Where’s the -> Operator?

In C and C++, two different operators are used for calling methods: you use . if you’re calling a method on the object directly and -> if you’re calling the method on a pointer to the object and need to dereference the pointer first. In other words, if object is a pointer, object->something() is similar to (*object).something().

Rust doesn’t have an equivalent to the -> operator; instead, Rust has a feature called automatic referencing and dereferencing. Calling methods is one of the few places in Rust that has this behavior.

Here’s how it works: when you call a method with object.something(), Rust automatically adds in &, &mut, or * so object matches the signature of the method. In other words, the following are the same:

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug,Copy,Clone)]
# struct Point {
#     x: f64,
#     y: f64,
# }
#
# impl Point {
#    fn distance(&self, other: &Point) -> f64 {
#        let x_squared = f64::powi(other.x - self.x, 2);
#        let y_squared = f64::powi(other.y - self.y, 2);
#
#        f64::sqrt(x_squared + y_squared)
#    }
# }
# let p1 = Point { x: 0.0, y: 0.0 };
# let p2 = Point { x: 5.0, y: 6.5 };
p1.distance(&p2);
(&p1).distance(&p2);
#}

The first one looks much cleaner. This automatic referencing behavior works because methods have a clear receiver—the type of self. Given the receiver and name of a method, Rust can figure out definitively whether the method is reading (&self), mutating (&mut self), or consuming (self). The fact that Rust makes borrowing implicit for method receivers is a big part of making ownership ergonomic in practice.

Let’s practice using methods by implementing a second method on the Rectangle struct. This time, we want an instance of Rectangle to take another instance of Rectangle and return true if the second Rectangle can fit completely within self; otherwise it should return false. That is, we want to be able to write the program shown in Listing 5-14, once we’ve defined the can_hold method:

Filename: src/main.rs

fn main() {
    let rect1 = Rectangle { width: 30, height: 50 };
    let rect2 = Rectangle { width: 10, height: 40 };
    let rect3 = Rectangle { width: 60, height: 45 };

    println!("Can rect1 hold rect2? {}", rect1.can_hold(&rect2));
    println!("Can rect1 hold rect3? {}", rect1.can_hold(&rect3));
}

Listing 5-14: Using the as-yet-unwritten can_hold method

And the expected output would look like the following, because both dimensions of rect2 are smaller than the dimensions of rect1 but rect3 is wider than rect1:

Can rect1 hold rect2? true
Can rect1 hold rect3? false

We know we want to define a method, so it will be within the impl Rectangle block. The method name will be can_hold, and it will take an immutable borrow of another Rectangle as a parameter. We can tell what the type of the parameter will be by looking at the code that calls the method: rect1.can_hold(&rect2) passes in &rect2, which is an immutable borrow to rect2, an instance of Rectangle. This makes sense because we only need to read rect2 (rather than write, which would mean we’d need a mutable borrow), and we want main to retain ownership of rect2 so we can use it again after calling the can_hold method. The return value of can_hold will be a Boolean, and the implementation will check whether the width and height of self are both greater than the width and height of the other Rectangle, respectively. Let’s add the new can_hold method to the impl block from Listing 5-13, shown in Listing 5-15:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# struct Rectangle {
#     width: u32,
#     height: u32,
# }
#
impl Rectangle {
    fn area(&self) -> u32 {
        self.width * self.height
    }

    fn can_hold(&self, other: &Rectangle) -> bool {
        self.width > other.width && self.height > other.height
    }
}
#}

Listing 5-15: Implementing the can_hold method on Rectangle that takes another Rectangle instance as a parameter

When we run this code with the main function in Listing 5-14, we’ll get our desired output. Methods can take multiple parameters that we add to the signature after the self parameter, and those parameters work just like parameters in functions.

Another useful feature of impl blocks is that we’re allowed to define functions within impl blocks that don’t take self as a parameter. These are called associated functions because they’re associated with the struct. They’re still functions, not methods, because they don’t have an instance of the struct to work with. You’ve already used the String::from associated function.

Associated functions are often used for constructors that will return a new instance of the struct. For example, we could provide an associated function that would have one dimension parameter and use that as both width and height, thus making it easier to create a square Rectangle rather than having to specify the same value twice:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# struct Rectangle {
#     width: u32,
#     height: u32,
# }
#
impl Rectangle {
    fn square(size: u32) -> Rectangle {
        Rectangle { width: size, height: size }
    }
}
#}

To call this associated function, we use the :: syntax with the struct name; let sq = Rectangle::square(3); is an example. This function is namespaced by the struct: the :: syntax is used for both associated functions and namespaces created by modules. We’ll discuss modules in Chapter 7.

Each struct is allowed to have multiple impl blocks. For example, Listing 5-15 is equivalent to the code shown in Listing 5-16, which has each method in its own impl block:

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# struct Rectangle {
#     width: u32,
#     height: u32,
# }
#
impl Rectangle {
    fn area(&self) -> u32 {
        self.width * self.height
    }
}

impl Rectangle {
    fn can_hold(&self, other: &Rectangle) -> bool {
        self.width > other.width && self.height > other.height
    }
}
#}

Listing 5-16: Rewriting Listing 5-15 using multiple impl blocks

There’s no reason to separate these methods into multiple impl blocks here, but this is valid syntax. We’ll see a case in which multiple impl blocks are useful in Chapter 10 where we discuss generic types and traits.

Structs let you create custom types that are meaningful for your domain. By using structs, you can keep associated pieces of data connected to each other and name each piece to make your code clear. Methods let you specify the behavior that instances of your structs have, and associated functions let you namespace functionality that is particular to your struct without having an instance available.

But structs aren’t the only way you can create custom types: let’s turn to Rust’s enum feature to add another tool to your toolbox.

In this chapter we’ll look at enumerations, also referred to as enums. Enums allow you to define a type by enumerating its possible values. First, we’ll define and use an enum to show how an enum can encode meaning along with data. Next, we’ll explore a particularly useful enum, called Option, which expresses that a value can be either something or nothing. Then we’ll look at how pattern matching in the match expression makes it easy to run different code for different values of an enum. Finally, we’ll cover how the if let construct is another convenient and concise idiom available to you to handle enums in your code.

Enums are a feature in many languages, but their capabilities differ in each language. Rust’s enums are most similar to algebraic data types in functional languages, such as F#, OCaml, and Haskell.

Let’s look at a situation we might want to express in code and see why enums are useful and more appropriate than structs in this case. Say we need to work with IP addresses. Currently, two major standards are used for IP addresses: version four and version six. These are the only possibilities for an IP address that our program will come across: we can enumerate all possible values, which is where enumeration gets its name.

Any IP address can be either a version four or a version six address, but not both at the same time. That property of IP addresses makes the enum data structure appropriate, because enum values can only be one of the variants. Both version four and version six addresses are still fundamentally IP addresses, so they should be treated as the same type when the code is handling situations that apply to any kind of IP address.

We can express this concept in code by defining an IpAddrKind enumeration and listing the possible kinds an IP address can be, V4 and V6. These are known as the variants of the enum:

# #![allow(unused_variables)]
#fn main() {
enum IpAddrKind {
    V4,
    V6,
}
#}

IpAddrKind is now a custom data type that we can use elsewhere in our code.

We can create instances of each of the two variants of IpAddrKind like this:

# #![allow(unused_variables)]
#fn main() {
# enum IpAddrKind {
#     V4,
#     V6,
# }
#
let four = IpAddrKind::V4;
let six = IpAddrKind::V6;
#}

Note that the variants of the enum are namespaced under its identifier, and we use a double colon to separate the two. The reason this is useful is that now both values IpAddrKind::V4 and IpAddrKind::V6 are of the same type: IpAddrKind. We can then, for instance, define a function that takes any IpAddrKind:

# #![allow(unused_variables)]
#fn main() {
# enum IpAddrKind {
#     V4,
#     V6,
# }
#
fn route(ip_type: IpAddrKind) { }
#}

And we can call this function with either variant:

# #![allow(unused_variables)]
#fn main() {
# enum IpAddrKind {
#     V4,
#     V6,
# }
#
# fn route(ip_type: IpAddrKind) { }
#
route(IpAddrKind::V4);
route(IpAddrKind::V6);
#}

Using enums has even more advantages. Thinking more about our IP address type, at the moment we don’t have a way to store the actual IP address data; we only know what kind it is. Given that you just learned about structs in Chapter 5, you might tackle this problem as shown in Listing 6-1:

# #![allow(unused_variables)]
#fn main() {
enum IpAddrKind {
    V4,
    V6,
}

struct IpAddr {
    kind: IpAddrKind,
    address: String,
}

let home = IpAddr {
    kind: IpAddrKind::V4,
    address: String::from("127.0.0.1"),
};

let loopback = IpAddr {
    kind: IpAddrKind::V6,
    address: String::from("::1"),
};
#}

Listing 6-1: Storing the data and IpAddrKind variant of an IP address using a struct

Here, we’ve defined a struct IpAddr that has two fields: a kind field that is of type IpAddrKind (the enum we defined previously) and an address field of type String. We have two instances of this struct. The first, home, has the value IpAddrKind::V4 as its kind with associated address data of 127.0.0.1. The second instance, loopback, has the other variant of IpAddrKind as its kind value, V6, and has address ::1 associated with it. We’ve used a struct to bundle the kind and address values together, so now the variant is associated with the value.

We can represent the same concept in a more concise way using just an enum, rather than an enum inside a struct, by putting data directly into each enum variant. This new definition of the IpAddr enum says that both V4 and V6 variants will have associated String values:

# #![allow(unused_variables)]
#fn main() {
enum IpAddr {
    V4(String),
    V6(String),
}

let home = IpAddr::V4(String::from("127.0.0.1"));

let loopback = IpAddr::V6(String::from("::1"));
#}

We attach data to each variant of the enum directly, so there is no need for an extra struct.

There’s another advantage to using an enum rather than a struct: each variant can have different types and amounts of associated data. Version four type IP addresses will always have four numeric components that will have values between 0 and 255. If we wanted to store V4 addresses as four u8 values but still express V6 addresses as one String value, we wouldn’t be able to with a struct. Enums handle this case with ease:

# #![allow(unused_variables)]
#fn main() {
enum IpAddr {
    V4(u8, u8, u8, u8),
    V6(String),
}

let home = IpAddr::V4(127, 0, 0, 1);

let loopback = IpAddr::V6(String::from("::1"));
#}

We’ve shown several different ways to define data structures to store version four and version six IP addresses. However, as it turns out, wanting to store IP addresses and encode which kind they are is so common that the standard library has a definition we can use! Let’s look at how the standard library defines IpAddr: it has the exact enum and variants that we’ve defined and used, but it embeds the address data inside the variants in the form of two different structs, which are defined differently for each variant:

# #![allow(unused_variables)]
#fn main() {
struct Ipv4Addr {
    // --snip--
}

struct Ipv6Addr {
    // --snip--
}

enum IpAddr {
    V4(Ipv4Addr),
    V6(Ipv6Addr),
}
#}

This code illustrates that you can put any kind of data inside an enum variant: strings, numeric types, or structs, for example. You can even include another enum! Also, standard library types are often not much more complicated than what you might come up with.

Note that even though the standard library contains a definition for IpAddr, we can still create and use our own definition without conflict because we haven’t brought the standard library’s definition into our scope. We’ll talk more about bringing types into scope in Chapter 7.

Let’s look at another example of an enum in Listing 6-2: this one has a wide variety of types embedded in its variants:

# #![allow(unused_variables)]
#fn main() {
enum Message {
    Quit,
    Move { x: i32, y: i32 },
    Write(String),
    ChangeColor(i32, i32, i32),
}
#}

Listing 6-2: A Message enum whose variants each store different amounts and types of values

This enum has four variants with different types:

• Quit has no data associated with it at all.
• Move includes an anonymous struct inside it.
• Write includes a single String.
• ChangeColor includes three i32 values.

Defining an enum with variants like the ones in Listing 6-2 is similar to defining different kinds of struct definitions, except the enum doesn’t use the struct keyword and all the variants are grouped together under the Message type. The following structs could hold the same data that the preceding enum variants hold:

# #![allow(unused_variables)]
#fn main() {
struct QuitMessage; // unit struct
struct MoveMessage {
    x: i32,
    y: i32,
}
struct WriteMessage(String); // tuple struct
struct ChangeColorMessage(i32, i32, i32); // tuple struct
#}

But if we used the different structs, which each have their own type, we couldn’t as easily define a function to take any of these kinds of messages as we could with the Message enum defined in Listing 6-2, which is a single type.

There is one more similarity between enums and structs: just as we’re able to define methods on structs using impl, we’re also able to define methods on enums. Here’s a method named call that we could define on our Message enum:

# #![allow(unused_variables)]
#fn main() {
# enum Message {
#     Quit,
#     Move { x: i32, y: i32 },
#     Write(String),
#     ChangeColor(i32, i32, i32),
# }
#
impl Message {
    fn call(&self) {
        // method body would be defined here
    }
}

let m = Message::Write(String::from("hello"));
m.call();
#}

The body of the method would use self to get the value that we called the method on. In this example, we’ve created a variable m that has the value Message::Write(String::from("hello")), and that is what self will be in the body of the call method when m.call() runs.

Let’s look at another enum in the standard library that is very common and useful: Option.

In the previous section, we looked at how the IpAddr enum let us use Rust’s type system to encode more information than just the data into our program. This section explores a case study of Option, which is another enum defined by the standard library. The Option type is used in many places because it encodes the very common scenario in which a value could be something or it could be nothing. Expressing this concept in terms of the type system means the compiler can check whether you’ve handled all the cases you should be handling; this functionality can prevent bugs that are extremely common in other programming languages.

Programming language design is often thought of in terms of which features you include, but the features you exclude are important too. Rust doesn’t have the null feature that many other languages have. Null is a value that means there is no value there. In languages with null, variables can always be in one of two states: null or not-null.

In his 2009 presentation “Null References: The Billion Dollar Mistake,” Tony Hoare, the inventor of null, has this to say:

I call it my billion-dollar mistake. At that time, I was designing the first comprehensive type system for references in an object-oriented language. My goal was to ensure that all use of references should be absolutely safe, with checking performed automatically by the compiler. But I couldn’t resist the temptation to put in a null reference, simply because it was so easy to implement. This has led to innumerable errors, vulnerabilities, and system crashes, which have probably caused a billion dollars of pain and damage in the last forty years.

The problem with null values is that if you try to use a null value as a not-null value, you’ll get an error of some kind. Because this null or not-null property is pervasive, it’s extremely easy to make this kind of error.

However, the concept that null is trying to express is still a useful one: a null is a value that is currently invalid or absent for some reason.

The problem isn’t really with the concept but with the particular implementation. As such, Rust does not have nulls, but it does have an enum that can encode the concept of a value being present or absent. This enum is Option<T>, and it is defined by the standard library as follows:

# #![allow(unused_variables)]
#fn main() {
enum Option<T> {
    Some(T),
    None,
}
#}

The Option<T> enum is so useful that it’s even included in the prelude; you don’t need to bring it into scope explicitly. In addition, so are its variants: you can use Some and None directly without the Option:: prefix. The Option<T> enum is still just a regular enum, and Some(T) and None are still variants of type Option<T>.

The <T> syntax is a feature of Rust we haven’t talked about yet. It’s a generic type parameter, and we’ll cover generics in more detail in Chapter 10. For now, all you need to know is that <T> means the Some variant of the Option enum can hold one piece of data of any type. Here are some examples of using Option values to hold number types and string types:

# #![allow(unused_variables)]
#fn main() {
let some_number = Some(5);
let some_string = Some("a string");

let absent_number: Option<i32> = None;
#}

If we use None rather than Some, we need to tell Rust what type of Option<T> we have, because the compiler can’t infer the type that the Some variant will hold by looking only at a None value.

When we have a Some value, we know that a value is present and the value is held within the Some. When we have a None value, in some sense, it means the same thing as null: we don’t have a valid value. So why is having Option<T> any better than having null?

In short, because Option<T> and T (where T can be any type) are different types, the compiler won’t let us use an Option<T> value as if it were definitely a valid value. For example, this code won’t compile because it’s trying to add an i8 to an Option<i8>:

let x: i8 = 5;
let y: Option<i8> = Some(5);

let sum = x + y;

If we run this code, we get an error message like this:

error[E0277]: the trait bound `i8: std::ops::Add<std::option::Option<i8>>` is
not satisfied
 -->
  |
5 |     let sum = x + y;
  |                 ^ no implementation for `i8 + std::option::Option<i8>`
  |

Intense! In effect, this error message means that Rust doesn’t understand how to add an i8 and an Option<i8>, because they’re different types. When we have a value of a type like i8 in Rust, the compiler will ensure that we always have a valid value. We can proceed confidently without having to check for null before using that value. Only when we have an Option<i8> (or whatever type of value we’re working with) do we have to worry about possibly not having a value, and the compiler will make sure we handle that case before using the value.

In other words, you have to convert an Option<T> to a T before you can perform T operations with it. Generally, this helps catch one of the most common issues with null: assuming that something isn’t null when it actually is.

Not having to worry about incorrectly assuming a not-null value helps you to be more confident in your code. In order to have a value that can possibly be null, you must explicitly opt in by making the type of that value Option<T>. Then, when you use that value, you are required to explicitly handle the case when the value is null. Everywhere that a value has a type that isn’t an Option<T>, you can safely assume that the value isn’t null. This was a deliberate design decision for Rust to limit null’s pervasiveness and increase the safety of Rust code.

So, how do you get the T value out of a Some variant when you have a value of type Option<T> so you can use that value? The Option<T> enum has a large number of methods that are useful in a variety of situations; you can check them out in its documentation. Becoming familiar with the methods on Option<T> will be extremely useful in your journey with Rust.

In general, in order to use an Option<T> value, you want to have code that will handle each variant. You want some code that will run only when you have a Some(T) value, and this code is allowed to use the inner T. You want some other code to run if you have a None value, and that code doesn’t have a T value available. The match expression is a control flow construct that does just this when used with enums: it will run different code depending on which variant of the enum it has, and that code can use the data inside the matching value.

Rust has an extremely powerful control flow operator called match that allows you to compare a value against a series of patterns and then execute code based on which pattern matches. Patterns can be made up of literal values, variable names, wildcards, and many other things; Chapter 18 covers all the different kinds of patterns and what they do. The power of match comes from the expressiveness of the patterns and the fact that the compiler confirms that all possible cases are handled.

Think of a match expression as being like a coin-sorting machine: coins slide down a track with variously sized holes along it, and each coin falls through the first hole it encounters that it fits into. In the same way, values go through each pattern in a match, and at the first pattern the value “fits,” the value falls into the associated code block to be used during execution.

Because we just mentioned coins, let’s use them as an example using match! We can write a function that can take an unknown United States coin and, in a similar way as the counting machine, determine which coin it is and return its value in cents, as shown here in Listing 6-3:

# #![allow(unused_variables)]
#fn main() {
enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter,
}

fn value_in_cents(coin: Coin) -> u32 {
    match coin {
        Coin::Penny => 1,
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter => 25,
    }
}
#}

Listing 6-3: An enum and a match expression that has the variants of the enum as its patterns

Let’s break down the match in the value_in_cents function. First, we list the match keyword followed by an expression, which in this case is the value coin. This seems very similar to an expression used with if, but there’s a big difference: with if, the expression needs to return a Boolean value, but here, it can be any type. The type of coin in this example is the Coin enum that we defined on line 1.

Next are the match arms. An arm has two parts: a pattern and some code. The first arm here has a pattern that is the value Coin::Penny and then the => operator that separates the pattern and the code to run. The code in this case is just the value 1. Each arm is separated from the next with a comma.

When the match expression executes, it compares the resulting value against the pattern of each arm, in order. If a pattern matches the value, the code associated with that pattern is executed. If that pattern doesn’t match the value, execution continues to the next arm, much as in a coin-sorting machine. We can have as many arms as we need: in Listing 6-3, our match has four arms.

The code associated with each arm is an expression, and the resulting value of the expression in the matching arm is the value that gets returned for the entire match expression.

Curly brackets typically aren’t used if the match arm code is short, as it is in Listing 6-3 where each arm just returns a value. If you want to run multiple lines of code in a match arm, you can use curly brackets. For example, the following code would print “Lucky penny!” every time the method was called with a Coin::Penny but would still return the last value of the block, 1:

# #![allow(unused_variables)]
#fn main() {
# enum Coin {
#    Penny,
#    Nickel,
#    Dime,
#    Quarter,
# }
#
fn value_in_cents(coin: Coin) -> u32 {
    match coin {
        Coin::Penny => {
            println!("Lucky penny!");
            1
        },
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter => 25,
    }
}
#}

Another useful feature of match arms is that they can bind to the parts of the values that match the pattern. This is how we can extract values out of enum variants.

As an example, let’s change one of our enum variants to hold data inside it. From 1999 through 2008, the United States minted quarters with different designs for each of the 50 states on one side. No other coins got state designs, so only quarters have this extra value. We can add this information to our enum by changing the Quarter variant to include a UsState value stored inside it, which we’ve done here in Listing 6-4:

# #![allow(unused_variables)]
#fn main() {
#[derive(Debug)] // So we can inspect the state in a minute
enum UsState {
    Alabama,
    Alaska,
    // --snip--
}

enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter(UsState),
}
#}

Listing 6-4: A Coin enum in which the Quarter variant also holds a UsState value

Let’s imagine that a friend of ours is trying to collect all 50 state quarters. While we sort our loose change by coin type, we’ll also call out the name of the state associated with each quarter so if it’s one our friend doesn’t have, they can add it to their collection.

In the match expression for this code, we add a variable called state to the pattern that matches values of the variant Coin::Quarter. When a Coin::Quarter matches, the state variable will bind to the value of that quarter’s state. Then we can use state in the code for that arm, like so:

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# enum UsState {
#    Alabama,
#    Alaska,
# }
#
# enum Coin {
#    Penny,
#    Nickel,
#    Dime,
#    Quarter(UsState),
# }
#
fn value_in_cents(coin: Coin) -> u32 {
    match coin {
        Coin::Penny => 1,
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter(state) => {
            println!("State quarter from {:?}!", state);
            25
        },
    }
}
#}

If we were to call value_in_cents(Coin::Quarter(UsState::Alaska)), coin would be Coin::Quarter(UsState::Alaska). When we compare that value with each of the match arms, none of them match until we reach Coin::Quarter(state). At that point, the binding for state will be the value UsState::Alaska. We can then use that binding in the println! expression, thus getting the inner state value out of the Coin enum variant for Quarter.

In the previous section, we wanted to get the inner T value out of the Some case when using Option<T>; we can also handle Option<T> using match as we did with the Coin enum! Instead of comparing coins, we’ll compare the variants of Option<T>, but the way that the match expression works remains the same.

Let’s say we want to write a function that takes an Option<i32> and, if there’s a value inside, adds 1 to that value. If there isn’t a value inside, the function should return the None value and not attempt to perform any operations.

This function is very easy to write, thanks to match, and will look like Listing 6-5:

# #![allow(unused_variables)]
#fn main() {
fn plus_one(x: Option<i32>) -> Option<i32> {
    match x {
        None => None,
        Some(i) => Some(i + 1),
    }
}

let five = Some(5);
let six = plus_one(five);
let none = plus_one(None);
#}

Listing 6-5: A function that uses a match expression on an Option<i32>

Let’s examine the first execution of plus_one in more detail. When we call plus_one(five), the variable x in the body of plus_one will have the value Some(5). We then compare that against each match arm.

None => None,

The Some(5) value doesn’t match the pattern None, so we continue to the next arm.

Some(i) => Some(i + 1),

Does Some(5) match Some(i)? Why yes it does! We have the same variant. The i binds to the value contained in Some, so i takes the value 5. The code in the match arm is then executed, so we add 1 to the value of i and create a new Some value with our total 6 inside.

Now let’s consider the second call of plus_one in Listing 6-5, where x is None. We enter the match and compare to the first arm.

None => None,

It matches! There’s no value to add to, so the program stops and returns the None value on the right side of =>. Because the first arm matched, no other arms are compared.

Combining match and enums is useful in many situations. You’ll see this pattern a lot in Rust code: match against an enum, bind a variable to the data inside, and then execute code based on it. It’s a bit tricky at first, but once you get used to it, you’ll wish you had it in all languages. It’s consistently a user favorite.

There’s one other aspect of match we need to discuss. Consider this version of our plus_one function that has a bug and won’t compile:

fn plus_one(x: Option<i32>) -> Option<i32> {
    match x {
        Some(i) => Some(i + 1),
    }
}

We didn’t handle the None case, so this code will cause a bug. Luckily, it’s a bug Rust knows how to catch. If we try to compile this code, we’ll get this error:

error[E0004]: non-exhaustive patterns: `None` not covered
 -->
  |
6 |         match x {
  |               ^ pattern `None` not covered

Rust knows that we didn’t cover every possible case and even knows which pattern we forgot! Matches in Rust are exhaustive: we must exhaust every last possibility in order for the code to be valid. Especially in the case of Option<T>, when Rust prevents us from forgetting to explicitly handle the None case, it protects us from assuming that we have a value when we might have null, thus making the billion-dollar mistake discussed earlier.

Rust also has a pattern we can use when we don’t want to list all possible values. For example, a u8 can have valid values of 0 through 255. If we only care about the values 1, 3, 5, and 7, we don’t want to have to list out 0, 2, 4, 6, 8, 9 all the way up to 255. Fortunately, we don’t have to: we can use the special pattern _ instead:

# #![allow(unused_variables)]
#fn main() {
let some_u8_value = 0u8;
match some_u8_value {
    1 => println!("one"),
    3 => println!("three"),
    5 => println!("five"),
    7 => println!("seven"),
    _ => (),
}
#}

The _ pattern will match any value. By putting it after our other arms, the _ will match all the possible cases that aren’t specified before it. The () is just the unit value, so nothing will happen in the _ case. As a result, we can say that we want to do nothing for all the possible values that we don’t list before the _ placeholder.

However, the match expression can be a bit wordy in a situation in which we only care about one of the cases. For this situation, Rust provides if let.

The if let syntax lets you combine if and let into a less verbose way to handle values that match one pattern while ignoring the rest. Consider the program in Listing 6-6 that matches on an Option<u8> value but only wants to execute code if the value is 3:

# #![allow(unused_variables)]
#fn main() {
let some_u8_value = Some(0u8);
match some_u8_value {
    Some(3) => println!("three"),
    _ => (),
}
#}

Listing 6-6: A match that only cares about executing code when the value is Some(3)

We want to do something with the Some(3) match but do nothing with any other Some<u8> value or the None value. To satisfy the match expression, we have to add _ => () after processing just one variant, which is a lot of boilerplate code to add.

Instead, we could write this in a shorter way using if let. The following code behaves the same as the match in Listing 6-6:

# #![allow(unused_variables)]
#fn main() {
# let some_u8_value = Some(0u8);
if let Some(3) = some_u8_value {
    println!("three");
}
#}

The syntax if let takes a pattern and an expression separated by an =. It works the same way as a match, where the expression is given to the match and the pattern is its first arm.

Using if let means you have less typing, less indentation, and less boilerplate code. However, you lose the exhaustive checking that match enforces. Choosing between match and if let depends on what you’re doing in your particular situation and whether gaining conciseness is an appropriate trade-off for losing exhaustive checking.

In other words, you can think of if let as syntax sugar for a match that runs code when the value matches one pattern and then ignores all other values.

We can include an else with an if let. The block of code that goes with the else is the same as the block of code that would go with the _ case in the match expression that is equivalent to the if let and else. Recall the Coin enum definition in Listing 6-4, where the Quarter variant also held a UsState value. If we wanted to count all non-quarter coins we see while also announcing the state of the quarters, we could do that with a match expression like this:

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# enum UsState {
#    Alabama,
#    Alaska,
# }
#
# enum Coin {
#    Penny,
#    Nickel,
#    Dime,
#    Quarter(UsState),
# }
# let coin = Coin::Penny;
let mut count = 0;
match coin {
    Coin::Quarter(state) => println!("State quarter from {:?}!", state),
    _ => count += 1,
}
#}

Or we could use an if let and else expression like this:

# #![allow(unused_variables)]
#fn main() {
# #[derive(Debug)]
# enum UsState {
#    Alabama,
#    Alaska,
# }
#
# enum Coin {
#    Penny,
#    Nickel,
#    Dime,
#    Quarter(UsState),
# }
# let coin = Coin::Penny;
let mut count = 0;
if let Coin::Quarter(state) = coin {
    println!("State quarter from {:?}!", state);
} else {
    count += 1;
}
#}

If you have a situation in which your program has logic that is too verbose to express using a match, remember that if let is in your Rust toolbox as well.

We’ve now covered how to use enums to create custom types that can be one of a set of enumerated values. We’ve shown how the standard library’s Option<T> type helps you use the type system to prevent errors. When enum values have data inside them, you can use match or if let to extract and use those values, depending on how many cases you need to handle.

Your Rust programs can now express concepts in your domain using structs and enums. Creating custom types to use in your API ensures type safety: the compiler will make certain your functions get only values of the type each function expects.

In order to provide a well-organized API to your users that is straightforward to use and only exposes exactly what your users will need, let’s now turn to Rust’s modules.

When you start writing programs in Rust, your code might live solely in the main function. As your code grows, you’ll eventually move functionality into other functions for reuse and better organization. By splitting your code into smaller chunks, you make each chunk easier to understand on its own. But what happens if you have too many functions? Rust has a module system that enables the reuse of code in an organized fashion.

In the same way that you extract lines of code into a function, you can extract functions (and other code, like structs and enums) into different modules. A module is a namespace that contains definitions of functions or types, and you can choose whether those definitions are visible outside their module (public) or not (private). Here’s an overview of how modules work:

• The mod keyword declares a new module. Code within the module appears either immediately following this declaration within curly brackets or in another file.
• By default, functions, types, constants, and modules are private. The pub keyword makes an item public and therefore visible outside its namespace.
• The use keyword brings modules, or the definitions inside modules, into scope so it’s easier to refer to them.

We’ll look at each of these parts to see how they fit into the whole.

We’ll start our module example by making a new project with Cargo, but instead of creating a binary crate, we’ll make a library crate: a project that other people can pull into their projects as a dependency. For example, the rand crate discussed in Chapter 2 is a library crate that we used as a dependency in the guessing game project.

We’ll create a skeleton of a library that provides some general networking functionality; we’ll concentrate on the organization of the modules and functions, but we won’t worry about what code goes in the function bodies. We’ll call our library communicator. To create a library, pass the --lib option instead of --bin:

$ cargo new communicator --lib
$ cd communicator

Notice that Cargo generated src/lib.rs instead of src/main.rs. Inside src/lib.rs we’ll find the following:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
    }
}
#}

Cargo creates an example test to help us get our library started, rather than the “Hello, world!” binary that we get when we use the --bin option. We’ll look at the #[] and mod tests syntax in the “Using super to Access a Parent Module” section later in this chapter, but for now, leave this code at the bottom of src/lib.rs.

Because we don’t have a src/main.rs file, there’s nothing for Cargo to execute with the cargo run command. Therefore, we’ll use the cargo build command to compile our library crate’s code.

We’ll look at different options for organizing your library’s code that will be suitable in a variety of situations, depending on the intent of the code.

For our communicator networking library, we’ll first define a module named network that contains the definition of a function called connect. Every module definition in Rust starts with the mod keyword. Add this code to the beginning of the src/lib.rs file, above the test code:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
mod network {
    fn connect() {
    }
}
#}

After the mod keyword, we put the name of the module, network, and then a block of code in curly brackets. Everything inside this block is inside the namespace network. In this case, we have a single function, connect. If we wanted to call this function from code outside the network module, we would need to specify the module and use the namespace syntax :: like so: network::connect().

We can also have multiple modules, side by side, in the same src/lib.rs file. For example, to also have a client module that has a function named connect, we can add it as shown in Listing 7-1:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
mod network {
    fn connect() {
    }
}

mod client {
    fn connect() {
    }
}
#}

Listing 7-1: The network module and the client module defined side by side in src/lib.rs

Now we have a network::connect function and a client::connect function. These can have completely different functionality, and the function names do not conflict with each other because they’re in different modules.

In this case, because we’re building a library, the file that serves as the entry point for building our library is src/lib.rs. However, in respect to creating modules, there’s nothing special about src/lib.rs. We could also create modules in src/main.rs for a binary crate in the same way as we’re creating modules in src/lib.rs for the library crate. In fact, we can put modules inside of modules, which can be useful as your modules grow to keep related functionality organized together and separate functionality apart. The way you choose to organize your code depends on how you think about the relationship between the parts of your code. For instance, the client code and its connect function might make more sense to users of our library if they were inside the network namespace instead, as in Listing 7-2:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
mod network {
    fn connect() {
    }

    mod client {
        fn connect() {
        }
    }
}
#}

Listing 7-2: Moving the client module inside the network module

In your src/lib.rs file, replace the existing mod network and mod client definitions with the ones in Listing 7-2, which have the client module as an inner module of network. The functions network::connect and network::client::connect are both named connect, but they don’t conflict with each other because they’re in different namespaces.

In this way, modules form a hierarchy. The contents of src/lib.rs are at the topmost level, and the submodules are at lower levels. Here’s what the organization of our example in Listing 7-1 looks like when thought of as a hierarchy:

communicator
 ├── network
 └── client

And here’s the hierarchy corresponding to the example in Listing 7-2:

communicator
 └── network
     └── client

The hierarchy shows that in Listing 7-2, client is a child of the network module rather than a sibling. More complicated projects can have many modules, and they’ll need to be organized logically in order for you to keep track of them. What “logically” means in your project is up to you and depends on how you and your library’s users think about your project’s domain. Use the techniques shown here to create side-by-side modules and nested modules in whatever structure you would like.

Modules form a hierarchical structure, much like another structure in computing that you’re used to: filesystems! We can use Rust’s module system along with multiple files to split up Rust projects so not everything lives in src/lib.rs or src/main.rs. For this example, let’s start with the code in Listing 7-3:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
mod client {
    fn connect() {
    }
}

mod network {
    fn connect() {
    }

    mod server {
        fn connect() {
        }
    }
}
#}

Listing 7-3: Three modules, client, network, and network::server, all defined in src/lib.rs

The file src/lib.rs has this module hierarchy:

communicator
 ├── client
 └── network
     └── server

If these modules had many functions, and those functions were becoming lengthy, it would be difficult to scroll through this file to find the code we wanted to work with. Because the functions are nested inside one or more mod blocks, the lines of code inside the functions will start getting lengthy as well. These would be good reasons to separate the client, network, and server modules from src/lib.rs and place them into their own files.

First, let’s replace the client module code with only the declaration of the client module so that src/lib.rs looks like code shown in Listing 7-4:

Filename: src/lib.rs

mod client;

mod network {
    fn connect() {
    }

    mod server {
        fn connect() {
        }
    }
}

Listing 7-4: Extracting the contents of the client module but leaving the declaration in src/lib.rs

We’re still declaring the client module here, but by replacing the block with a semicolon, we’re telling Rust to look in another location for the code defined within the scope of the client module. In other words, the line mod client; means this:

mod client {
    // contents of client.rs
}

Now we need to create the external file with that module name. Create a client.rs file in your src/ directory and open it. Then enter the following, which is the connect function in the client module that we removed in the previous step:

Filename: src/client.rs

# #![allow(unused_variables)]
#fn main() {
fn connect() {
}
#}

Note that we don’t need a mod declaration in this file because we already declared the client module with mod in src/lib.rs. This file just provides the contents of the client module. If we put a mod client here, we’d be giving the client module its own submodule named client!

Rust only knows to look in src/lib.rs by default. If we want to add more files to our project, we need to tell Rust in src/lib.rs to look in other files; this is why mod client needs to be defined in src/lib.rs and can’t be defined in src/client.rs.

Now the project should compile successfully, although you’ll get a few warnings. Remember to use cargo build instead of cargo run because we have a library crate rather than a binary crate:

$ cargo build
   Compiling communicator v0.1.0 (file:///projects/communicator)
warning: function is never used: `connect`
 --> src/client.rs:1:1
  |
1 | / fn connect() {
2 | | }
  | |_^
  |
  = note: #[warn(dead_code)] on by default

warning: function is never used: `connect`
 --> src/lib.rs:4:5
  |
4 | /     fn connect() {
5 | |     }
  | |_____^

warning: function is never used: `connect`
 --> src/lib.rs:8:9
  |
8 | /         fn connect() {
9 | |         }
  | |_________^

These warnings tell us that we have functions that are never used. Don’t worry about these warnings for now; we’ll address them later in this chapter in the “Controlling Visibility with pub” section. The good news is that they’re just warnings; our project built successfully!

Next, let’s extract the network module into its own file using the same pattern. In src/lib.rs, delete the body of the network module and add a semicolon to the declaration, like so:

Filename: src/lib.rs

mod client;

mod network;

Then create a new src/network.rs file and enter the following:

Filename: src/network.rs

# #![allow(unused_variables)]
#fn main() {
fn connect() {
}

mod server {
    fn connect() {
    }
}
#}

Notice that we still have a mod declaration within this module file; this is because we still want server to be a submodule of network.

Run cargo build again. Success! We have one more module to extract: server. Because it’s a submodule—that is, a module within a module—our current tactic of extracting a module into a file named after that module won’t work. We’ll try anyway so you can see the error. First, change src/network.rs to have mod server; instead of the server module’s contents:

Filename: src/network.rs

fn connect() {
}

mod server;

Then create a src/server.rs file and enter the contents of the server module that we extracted:

Filename: src/server.rs

# #![allow(unused_variables)]
#fn main() {
fn connect() {
}
#}

When we try to cargo build, we’ll get the error shown in Listing 7-5:

$ cargo build
   Compiling communicator v0.1.0 (file:///projects/communicator)
error: cannot declare a new module at this location
 --> src/network.rs:4:5
  |
4 | mod server;
  |     ^^^^^^
  |
note: maybe move this module `src/network.rs` to its own directory via `src/network/mod.rs`
 --> src/network.rs:4:5
  |
4 | mod server;
  |     ^^^^^^
note: ... or maybe `use` the module `server` instead of possibly redeclaring it
 --> src/network.rs:4:5
  |
4 | mod server;
  |     ^^^^^^

Listing 7-5: Error when trying to extract the server submodule into src/server.rs

The error says we cannot declare a new module at this location and is pointing to the mod server; line in src/network.rs. So src/network.rs is different than src/lib.rs somehow: keep reading to understand why.

The note in the middle of Listing 7-5 is actually very helpful because it points out something we haven’t yet talked about doing:

note: maybe move this module `network` to its own directory via
`network/mod.rs`

Instead of continuing to follow the same file-naming pattern we used previously, we can do what the note suggests:

1. Make a new directory named network, the parent module’s name.
2. Move the src/network.rs file into the new network directory and rename it src/network/mod.rs.
3. Move the submodule file src/server.rs into the network directory.

Here are commands to carry out these steps:

$ mkdir src/network
$ mv src/network.rs src/network/mod.rs
$ mv src/server.rs src/network

Now when we try to run cargo build, compilation will work (we’ll still have warnings though). Our module layout still looks exactly the same as it did when we had all the code in src/lib.rs in Listing 7-3:

communicator
 ├── client
 └── network
     └── server

The corresponding file layout now looks like this:

└── src
    ├── client.rs
    ├── lib.rs
    └── network
        ├── mod.rs
        └── server.rs

So when we wanted to extract the network::server module, why did we have to also change the src/network.rs file to the src/network/mod.rs file and put the code for network::server in the network directory in src/network/server.rs? Why couldn’t we just extract the network::server module into src/server.rs? The reason is that Rust wouldn’t be able to recognize that server was supposed to be a submodule of network if the server.rs file was in the src directory. To clarify Rust’s behavior here, let’s consider a different example with the following module hierarchy, where all the definitions are in src/lib.rs:

communicator
 ├── client
 └── network
     └── client

In this example, we have three modules again: client, network, and network::client. Following the same steps we did earlier for extracting modules into files, we would create src/client.rs for the client module. For the network module, we would create src/network.rs. But we wouldn’t be able to extract the network::client module into a src/client.rs file because that already exists for the top-level client module! If we could put the code for both the client and network::client modules in the src/client.rs file, Rust wouldn’t have any way to know whether the code was for client or for network::client.

Therefore, in order to extract a file for the network::client submodule of the network module, we needed to create a directory for the network module instead of a src/network.rs file. The code that is in the network module then goes into the src/network/mod.rs file, and the submodule network::client can have its own src/network/client.rs file. Now the top-level src/client.rs is unambiguously the code that belongs to the client module.

Let’s summarize the rules of modules with regard to files:

• If a module named foo has no submodules, you should put the declarations for foo in a file named foo.rs.
• If a module named foo does have submodules, you should put the declarations for foo in a file named foo/mod.rs.

These rules apply recursively, so if a module named foo has a submodule named bar and bar does not have submodules, you should have the following files in your src directory:

└── foo
    ├── bar.rs (contains the declarations in `foo::bar`)
    └── mod.rs (contains the declarations in `foo`, including `mod bar`)

The modules should be declared in their parent module’s file using the mod keyword.

Next, we’ll talk about the pub keyword and get rid of those warnings!

We resolved the error messages shown in Listing 7-5 by moving the network and network::server code into the src/network/mod.rs and src/network/server.rs files, respectively. At that point, cargo build was able to build our project, but we still get warning messages about the client::connect, network::connect, and network::server::connect functions not being used.

So why are we receiving these warnings? After all, we’re building a library with functions that are intended to be used by our users, not necessarily by us within our own project, so it shouldn’t matter that these connect functions go unused. The point of creating them is that they will be used by another project, not our own.

To understand why this program invokes these warnings, let’s try using the connect library from another project, calling it externally. To do that, we’ll create a binary crate in the same directory as our library crate by making a src/main.rs file containing this code:

Filename: src/main.rs

extern crate communicator;

fn main() {
    communicator::client::connect();
}

We use the extern crate command to bring the communicator library crate into scope. Our package now contains two crates. Cargo treats src/main.rs as the root file of a binary crate, which is separate from the existing library crate whose root file is src/lib.rs. This pattern is quite common for executable projects: most functionality is in a library crate, and the binary crate uses that library crate. As a result, other programs can also use the library crate, and it’s a nice separation of concerns.

From the point of view of a crate outside the communicator library looking in, all the modules we’ve been creating are within a module that has the same name as the crate, communicator. We call the top-level module of a crate the root module.

Also note that even if we’re using an external crate within a submodule of our project, the extern crate should go in our root module (so in src/main.rs or src/lib.rs). Then, in our submodules, we can refer to items from external crates as if the items are top-level modules.

Right now, our binary crate just calls our library’s connect function from the client module. However, invoking cargo build will now give us an error after the warnings:

error[E0603]: module `client` is private
 --> src/main.rs:4:5
  |
4 |     communicator::client::connect();
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Ah ha! This error tells us that the client module is private, which is the crux of the warnings. It’s also the first time we’ve run into the concepts of public and private in the context of Rust. The default state of all code in Rust is private: no one else is allowed to use the code. If you don’t use a private function within your program, because your program is the only code allowed to use that function, Rust will warn you that the function has gone unused.

After you specify that a function such as client::connect is public, not only will your call to that function from your binary crate be allowed, but also the warning that the function is unused will go away. Marking a function as public lets Rust know that the function will be used by code outside of your program. Rust considers the theoretical external usage that’s now possible as the function “being used.” Thus, when a function is marked public, Rust will not require that it be used in your program and will stop warning that the function is unused.

To tell Rust to make a function public, we add the pub keyword to the start of the declaration. We’ll focus on fixing the warning that indicates client::connect has gone unused for now, as well as the module `client` is private error from our binary crate. Modify src/lib.rs to make the client module public, like so:

Filename: src/lib.rs

pub mod client;

mod network;

The pub keyword is placed right before mod. Let’s try building again:

error[E0603]: function `connect` is private
 --> src/main.rs:4:5
  |
4 |     communicator::client::connect();
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Hooray! We have a different error! Yes, different error messages are a cause for celebration. The new error shows function `connect` is private, so let’s edit src/client.rs to make client::connect public too:

Filename: src/client.rs

# #![allow(unused_variables)]
#fn main() {
pub fn connect() {
}
#}

Now run cargo build again:

warning: function is never used: `connect`
 --> src/network/mod.rs:1:1
  |
1 | / fn connect() {
2 | | }
  | |_^
  |
  = note: #[warn(dead_code)] on by default

warning: function is never used: `connect`
 --> src/network/server.rs:1:1
  |
1 | / fn connect() {
2 | | }
  | |_^

The code compiled, and the warning that client::connect is not being used is gone!

Unused code warnings don’t always indicate that an item in your code needs to be made public: if you didn’t want these functions to be part of your public API, unused code warnings could be alerting you to code you no longer need that you can safely delete. They could also be alerting you to a bug if you had just accidentally removed all places within your library where this function is called.

But in this case, we do want the other two functions to be part of our crate’s public API, so let’s mark them as pub as well to get rid of the remaining warnings. Modify src/network/mod.rs to look like the following:

Filename: src/network/mod.rs

pub fn connect() {
}

mod server;

Then compile the code:

warning: function is never used: `connect`
 --> src/network/mod.rs:1:1
  |
1 | / pub fn connect() {
2 | | }
  | |_^
  |
  = note: #[warn(dead_code)] on by default

warning: function is never used: `connect`
 --> src/network/server.rs:1:1
  |
1 | / fn connect() {
2 | | }
  | |_^

Hmmm, we’re still getting an unused function warning, even though network::connect is set to pub. The reason is that the function is public within the module, but the network module that the function resides in is not public. We’re working from the interior of the library out this time, whereas with client::connect we worked from the outside in. We need to change src/lib.rs to make network public too, like so:

Filename: src/lib.rs

pub mod client;

pub mod network;

Now when we compile, that warning is gone:

warning: function is never used: `connect`
 --> src/network/server.rs:1:1
  |
1 | / fn connect() {
2 | | }
  | |_^
  |
  = note: #[warn(dead_code)] on by default

Only one warning is left—try to fix this one on your own!

Overall, these are the rules for item visibility:

• If an item is public, it can be accessed through any of its parent modules.
• If an item is private, it can be accessed only by its immediate parent module and any of the parent’s child modules.

Let’s look at a few more privacy examples to get some practice. Create a new library project and enter the code in Listing 7-6 into your new project’s src/lib.rs:

Filename: src/lib.rs

mod outermost {
    pub fn middle_function() {}

    fn middle_secret_function() {}

    mod inside {
        pub fn inner_function() {}

        fn secret_function() {}
    }
}

fn try_me() {
    outermost::middle_function();
    outermost::middle_secret_function();
    outermost::inside::inner_function();
    outermost::inside::secret_function();
}

Listing 7-6: Examples of private and public functions, some of which are incorrect

Before you try to compile this code, make a guess about which lines in the try_me function will have errors. Then, try compiling the code to see whether you were right—and read on for the discussion of the errors!

The try_me function is in the root module of our project. The module named outermost is private, but the second privacy rule states that the try_me function is allowed to access the outermost module because outermost is in the current (root) module, as is try_me.

The call to outermost::middle_function will work because middle_function is public and try_me is accessing middle_function through its parent module outermost. We determined in the previous paragraph that this module is accessible.

The call to outermost::middle_secret_function will cause a compilation error. Because middle_secret_function is private, the second rule applies. The root module is neither the current module of middle_secret_function (outermost is), nor is it a child module of the current module of middle_secret_function.

The module named inside is private and has no child modules, so it can be accessed only by its current module outermost. That means the try_me function is not allowed to call outermost::inside::inner_function or outermost::inside::secret_function.

Here are some suggestions for changing the code in an attempt to fix the errors. Make a guess as to whether it will fix the errors before you try each one. Then compile the code to see whether or not you’re right, using the privacy rules to understand why. Feel free to design more experiments and try them out!

• What if the inside module were public?
• What if outermost were public and inside were private?
• What if, in the body of inner_function, you called ::outermost::middle_secret_function()? (The two colons at the beginning mean that we want to refer to the modules starting from the root module.)

Next, let’s talk about bringing items into scope with the use keyword.

We’ve covered how to call functions defined within a module using the module name as part of the call, as in the call to the nested_modules function shown here in Listing 7-7:

Filename: src/main.rs

pub mod a {
    pub mod series {
        pub mod of {
            pub fn nested_modules() {}
        }
    }
}

fn main() {
    a::series::of::nested_modules();
}

Listing 7-7: Calling a function by fully specifying its enclosing module’s path

As you can see, referring to the fully qualified name can get quite lengthy. Fortunately, Rust has a keyword to make these calls more concise.

Rust’s use keyword shortens lengthy function calls by bringing the modules of the function you want to call into scope. Here’s an example of bringing the a::series::of module into a binary crate’s root scope:

Filename: src/main.rs

pub mod a {
    pub mod series {
        pub mod of {
            pub fn nested_modules() {}
        }
    }
}

use a::series::of;

fn main() {
    of::nested_modules();
}

The line use a::series::of; means that rather than using the full a::series::of path wherever we want to refer to the of module, we can use of.

The use keyword brings only what we’ve specified into scope: it does not bring children of modules into scope. That’s why we still have to use of::nested_modules when we want to call the nested_modules function.

We could have chosen to bring the function into scope by instead specifying the function in the use as follows:

pub mod a {
    pub mod series {
        pub mod of {
            pub fn nested_modules() {}
        }
    }
}

use a::series::of::nested_modules;

fn main() {
    nested_modules();
}

Doing so allows us to exclude all the modules and reference the function directly.

Because enums also form a sort of namespace like modules, we can bring an enum’s variants into scope with use as well. For any kind of use statement, if you’re bringing multiple items from one namespace into scope, you can list them using curly brackets and commas in the last position, like so:

enum TrafficLight {
    Red,
    Yellow,
    Green,
}

use TrafficLight::{Red, Yellow};

fn main() {
    let red = Red;
    let yellow = Yellow;
    let green = TrafficLight::Green;
}

We’re still specifying the TrafficLight namespace for the Green variant because we didn’t include Green in the use statement.

To bring all the items in a namespace into scope at once, we can use the * syntax, which is called the glob operator. This example brings all the variants of an enum into scope without having to list each specifically:

enum TrafficLight {
    Red,
    Yellow,
    Green,
}

use TrafficLight::*;

fn main() {
    let red = Red;
    let yellow = Yellow;
    let green = Green;
}

The * will bring into scope all the visible items in the TrafficLight namespace. You should use globs sparingly: they are convenient, but a glob might also pull in more items than you expected and cause naming conflicts.

As you saw at the beginning of this chapter, when you create a library crate, Cargo makes a tests module for you. Let’s go into more detail about that now. In your communicator project, open src/lib.rs:

Filename: src/lib.rs

pub mod client;

pub mod network;

#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
    }
}

Chapter 11 explains more about testing, but parts of this example should make sense now: we have a module named tests that lives next to our other modules and contains one function named it_works. Even though there are special annotations, the tests module is just another module! So our module hierarchy looks like this:

communicator
 ├── client
 ├── network
 |   └── client
 └── tests

Tests are for exercising the code within our library, so let’s try to call our client::connect function from this it_works function, even though we won’t be checking any functionality right now. This won’t work yet:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
        client::connect();
    }
}
#}

Run the tests by invoking the cargo test command:

$ cargo test
   Compiling communicator v0.1.0 (file:///projects/communicator)
error[E0433]: failed to resolve. Use of undeclared type or module `client`
 --> src/lib.rs:9:9
  |
9 |         client::connect();
  |         ^^^^^^ Use of undeclared type or module `client`

The compilation failed, but why? We don’t need to place communicator:: in front of the function, as we did in src/main.rs, because we are definitely within the communicator library crate here. The reason is that paths are always relative to the current module, which here is tests. The only exception is in a use statement, where paths are relative to the crate root by default. Our tests module needs the client module in its scope!

So how do we get back up one module in the module hierarchy to call the client::connect function in the tests module? In the tests module, we can either use leading colons to let Rust know that we want to start from the root and list the whole path, like this:

::client::connect();

Or, we can use super to move up one module in the hierarchy from our current module, like this:

super::client::connect();

These two options don’t look that different in this example, but if you’re deeper in a module hierarchy, starting from the root every time would make your code lengthy. In those cases, using super to get from the current module to sibling modules is a good shortcut. Plus, if you’ve specified the path from the root in many places in your code and then rearrange your modules by moving a subtree to another place, you’ll end up needing to update the path in several places, which would be tedious.

It would also be annoying to have to type super:: in each test, but you’ve already seen the tool for that solution: use! The super:: functionality changes the path you give to use so it is relative to the parent module instead of to the root module.

For these reasons, in the tests module especially, use super::something is usually the best solution. So now our test looks like this:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[cfg(test)]
mod tests {
    use super::client;

    #[test]
    fn it_works() {
        client::connect();
    }
}
#}

When we run cargo test again, the test will pass, and the first part of the test result output will be the following:

$ cargo test
   Compiling communicator v0.1.0 (file:///projects/communicator)
     Running target/debug/communicator-92007ddb5330fa5a

running 1 test
test tests::it_works ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Now you know some new techniques for organizing your code! Use these techniques to group related functionality together, keep files from becoming too long, and present a tidy public API to your library users.

Next, we’ll look at some collection data structures in the standard library that you can use in your nice, neat code.

Rust’s standard library includes a number of very useful data structures called collections. Most other data types represent one specific value, but collections can contain multiple values. Unlike the built-in array and tuple types, the data these collections point to is stored on the heap, which means the amount of data does not need to be known at compile time and can grow or shrink as the program runs. Each kind of collection has different capabilities and costs, and choosing an appropriate one for your current situation is a skill you’ll develop over time. In this chapter, we’ll discuss three collections that are used very often in Rust programs:

• A vector allows you to store a variable number of values next to each other.
• A string is a collection of characters. We’ve mentioned the String type previously, but in this chapter we’ll talk about it in depth.
• A hash map allows you to associate a value with a particular key. It’s a particular implementation of the more general data structure called a map.

To learn about the other kinds of collections provided by the standard library, see the documentation.

We’ll discuss how to create and update vectors, strings, and hash maps, as well as what makes each special.

The first collection type we’ll look at is Vec<T>, also known as a vector. Vectors allow you to store more than one value in a single data structure that puts all the values next to each other in memory. Vectors can only store values of the same type. They are useful when you have a list of items, such as the lines of text in a file or the prices of items in a shopping cart.

To create a new, empty vector, we can call the Vec::new function, as shown in Listing 8-1:

# #![allow(unused_variables)]
#fn main() {
let v: Vec<i32> = Vec::new();
#}

Listing 8-1: Creating a new, empty vector to hold values of type i32

Note that we added a type annotation here. Because we aren’t inserting any values into this vector, Rust doesn’t know what kind of elements we intend to store. This is an important point. Vectors are implemented using generics; we’ll cover how to use generics with your own types in Chapter 10. For now, know that the Vec<T> type provided by the standard library can hold any type, and when a specific vector holds a specific type, the type is specified within angle brackets. In Listing 8-1, we’ve told Rust that the Vec<T> in v will hold elements of the i32 type.

In more realistic code, Rust can often infer the type of value you want to store once you insert values, so you rarely need to do this type annotation. It’s more common to create a Vec<T> that has initial values, and Rust provides the vec! macro for convenience. The macro will create a new vector that holds the values you give it. Listing 8-2 creates a new Vec<i32> that holds the values 1, 2, and 3:

# #![allow(unused_variables)]
#fn main() {
let v = vec![1, 2, 3];
#}

Listing 8-2: Creating a new vector containing values

Because we’ve given initial i32 values, Rust can infer that the type of v is Vec<i32>, and the type annotation isn’t necessary. Next, we’ll look at how to modify a vector.

To create a vector and then add elements to it, we can use the push method, as shown in Listing 8-3:

# #![allow(unused_variables)]
#fn main() {
let mut v = Vec::new();

v.push(5);
v.push(6);
v.push(7);
v.push(8);
#}

Listing 8-3: Using the push method to add values to a vector

As with any variable, if we want to be able to change its value, we need to make it mutable using the mut keyword, as discussed in Chapter 3. The numbers we place inside are all of type i32, and Rust infers this from the data, so we don’t need the Vec<i32> annotation.

Like any other struct, a vector is freed when it goes out of scope, as annotated in Listing 8-4:

# #![allow(unused_variables)]
#fn main() {
{
    let v = vec![1, 2, 3, 4];

    // do stuff with v

} // <- v goes out of scope and is freed here
#}

Listing 8-4: Showing where the vector and its elements are dropped

When the vector gets dropped, all of its contents are also dropped, meaning those integers it holds will be cleaned up. This may seem like a straightforward point but can get a bit more complicated when you start to introduce references to the elements of the vector. Let’s tackle that next!

Now that you know how to create, update, and destroy vectors, knowing how to read their contents is a good next step. There are two ways to reference a value stored in a vector. In the examples, we’ve annotated the types of the values that are returned from these functions for extra clarity.

Listing 8-5 shows both methods of accessing a value in a vector, either with indexing syntax or the get method:

# #![allow(unused_variables)]
#fn main() {
let v = vec![1, 2, 3, 4, 5];

let third: &i32 = &v[2];
let third: Option<&i32> = v.get(2);
#}

Listing 8-5: Using indexing syntax or the get method to access an item in a vector

Note two details here. First, we use the index value of 2 to get the third element: vectors are indexed by number, starting at zero. Second, the two ways to get the third element are by using & and [], which gives us a reference, or by using the get method with the index passed as an argument, which gives us an Option<&T>.

Rust has two ways to reference an element so you can choose how the program behaves when you try to use an index value that the vector doesn’t have an element for. As an example, let’s see what a program will do if it has a vector that holds five elements and then tries to access an element at index 100, as shown in Listing 8-6:

# #![allow(unused_variables)]
#fn main() {
let v = vec![1, 2, 3, 4, 5];

let does_not_exist = &v[100];
let does_not_exist = v.get(100);
#}

Listing 8-6: Attempting to access the element at index 100 in a vector containing five elements

When we run this code, the first [] method will cause the program to panic because it references a nonexistent element. This method is best used when you want your program to crash if there’s an attempt to access an element past the end of the vector.

When the get method is passed an index that is outside the vector, it returns None without panicking. You would use this method if accessing an element beyond the range of the vector happens occasionally under normal circumstances. Your code will then have logic to handle having either Some(&element) or None, as discussed in Chapter 6. For example, the index could be coming from a person entering a number. If they accidentally enter a number that’s too large and the program gets a None value, you could tell the user how many items are in the current vector and give them another chance to enter a valid value. That would be more user-friendly than crashing the program due to a typo!

When the program has a valid reference, the borrow checker enforces the ownership and borrowing rules (covered in Chapter 4) to ensure this reference and any other references to the contents of the vector remain valid. Recall the rule that states you can’t have mutable and immutable references in the same scope. That rule applies in Listing 8-7, where we hold an immutable reference to the first element in a vector and try to add an element to the end, which won’t work:

let mut v = vec![1, 2, 3, 4, 5];

let first = &v[0];

v.push(6);

Listing 8-7: Attempting to add an element to a vector while holding a reference to an item

Compiling this code will result in this error:

error[E0502]: cannot borrow `v` as mutable because it is also borrowed as immutable
 -->
  |
4 |     let first = &v[0];
  |                  - immutable borrow occurs here
5 |
6 |     v.push(6);
  |     ^ mutable borrow occurs here
7 |
8 | }
  | - immutable borrow ends here

The code in Listing 8-7 might look like it should work: why should a reference to the first element care about what changes at the end of the vector? This error is due to the way vectors work: adding a new element onto the end of the vector might require allocating new memory and copying the old elements to the new space, if there isn’t enough room to put all the elements next to each other where the vector currently is. In that case, the reference to the first element would be pointing to deallocated memory. The borrowing rules prevent programs from ending up in that situation.

Note: For more on the implementation details of the Vec<T> type, see “The Rustonomicon” at https://doc.rust-lang.org/stable/nomicon/vec.html.

If we want to access each element in a vector in turn, we can iterate through all of the elements rather than use indexes to access one at a time. Listing 8-8 shows how to use a for loop to get immutable references to each element in a vector of i32 values and print them:

# #![allow(unused_variables)]
#fn main() {
let v = vec![100, 32, 57];
for i in &v {
    println!("{}", i);
}
#}

Listing 8-8: Printing each element in a vector by iterating over the elements using a for loop

We can also iterate over mutable references to each element in a mutable vector in order to make changes to all the elements. The for loop in Listing 8-9 will add 50 to each element:

# #![allow(unused_variables)]
#fn main() {
let mut v = vec![100, 32, 57];
for i in &mut v {
    *i += 50;
}
#}

Listing 8-9: Iterating over mutable references to elements in a vector

To change the value that the mutable reference refers to, we have to use the dereference operator (*) to get to the value in i before we can use the += operator .

At the beginning of this chapter, we said that vectors can only store values that are the same type. This can be inconvenient; there are definitely use cases for needing to store a list of items of different types. Fortunately, the variants of an enum are defined under the same enum type, so when we need to store elements of a different type in a vector, we can define and use an enum!

For example, say we want to get values from a row in a spreadsheet in which some of the columns in the row contain integers, some floating-point numbers, and some strings. We can define an enum whose variants will hold the different value types, and then all the enum variants will be considered the same type: that of the enum. Then we can create a vector that holds that enum and so, ultimately, holds different types. We’ve demonstrated this in Listing 8-10:

# #![allow(unused_variables)]
#fn main() {
enum SpreadsheetCell {
    Int(i32),
    Float(f64),
    Text(String),
}

let row = vec![
    SpreadsheetCell::Int(3),
    SpreadsheetCell::Text(String::from("blue")),
    SpreadsheetCell::Float(10.12),
];
#}

Listing 8-10: Defining an enum to store values of different types in one vector

Rust needs to know what types will be in the vector at compile time so it knows exactly how much memory on the heap will be needed to store each element. A secondary advantage is that we can be explicit about what types are allowed in this vector. If Rust allowed a vector to hold any type, there would be a chance that one or more of the types would cause errors with the operations performed on the elements of the vector. Using an enum plus a match expression means that Rust will ensure at compile time that every possible case is handled, as discussed in Chapter 6.

When you’re writing a program, if you don’t know the exhaustive set of types the program will get at runtime to store in a vector, the enum technique won’t work. Instead, you can use a trait object, which we’ll cover in Chapter 17.

Now that we’ve discussed some of the most common ways to use vectors, be sure to review the API documentation for all the many useful methods defined on Vec<T> by the standard library. For example, in addition to push, a pop method removes and returns the last element. Let’s move on to the next collection type: String!

We talked about strings in Chapter 4, but we’ll look at them in more depth now. New Rustaceans commonly get stuck on strings due to a combination of three reasons: Rust’s propensity for exposing possible errors, strings being a more complicated data structure than many programmers give them credit for, and UTF-8. These factors combine in a way that can seem difficult when you’re coming from other programming languages.

It’s useful to discuss strings in the context of collections because strings are implemented as a collection of bytes, plus some methods to provide useful functionality when those bytes are interpreted as text. In this section, we’ll talk about the operations on String that every collection type has, such as creating, updating, and reading. We’ll also discuss the ways in which String is different from the other collections, namely how indexing into a String is complicated by the differences between how people and computers interpret String data.

We’ll first define what we mean by the term string. Rust has only one string type in the core language, which is the string slice str that is usually seen in its borrowed form &str. In Chapter 4, we talked about string slices, which are references to some UTF-8 encoded string data stored elsewhere. String literals, for example, are stored in the binary output of the program and are therefore string slices.

The String type, which is provided by Rust’s standard library rather than coded into the core language, is a growable, mutable, owned, UTF-8 encoded string type. When Rustaceans refer to “strings” in Rust, they usually mean the String and the string slice &str types, not just one of those types. Although this section is largely about String, both types are used heavily in Rust’s standard library, and both String and string slices are UTF-8 encoded.

Rust’s standard library also includes a number of other string types, such as OsString, OsStr, CString, and CStr. Library crates can provide even more options for storing string data. See how those names all end in String or Str? They refer to owned and borrowed variants, just like the String and str types you’ve seen previously. These string types can store text in different encodings or be represented in memory in a different way, for example. We won’t discuss these other string types in this chapter; see their API documentation for more about how to use them and when each is appropriate.

Many of the same operations available with Vec<T> are available with String as well, starting with the new function to create a string, shown in Listing 8-11:

# #![allow(unused_variables)]
#fn main() {
let mut s = String::new();
#}

Listing 8-11: Creating a new, empty String

This line creates a new empty string called s, which we can then load data into. Often, we’ll have some initial data that we want to start the string with. For that, we use the to_string method, which is available on any type that implements the Display trait, as string literals do. Listing 8-12 shows two examples:

# #![allow(unused_variables)]
#fn main() {
let data = "initial contents";

let s = data.to_string();

// the method also works on a literal directly:
let s = "initial contents".to_string();
#}

Listing 8-12: Using the to_string method to create a String from a string literal

This code creates a string containing initial contents.

We can also use the function String::from to create a String from a string literal. The code in Listing 8-13 is equivalent to the code from Listing 8-12 that uses to_string:

# #![allow(unused_variables)]
#fn main() {
let s = String::from("initial contents");
#}

Listing 8-13: Using the String::from function to create a String from a string literal

Because strings are used for so many things, we can use many different generic APIs for strings, providing us with a lot of options. Some of them can seem redundant, but they all have their place! In this case, String::from and to_string do the same thing, so which you choose is a matter of style.

Remember that strings are UTF-8 encoded, so we can include any properly encoded data in them, as shown in Listing 8-14:

# #![allow(unused_variables)]
#fn main() {
let hello = String::from("السلام عليكم");
let hello = String::from("Dobrý den");
let hello = String::from("Hello");
let hello = String::from("שָׁלוֹם");
let hello = String::from("नमस्ते");
let hello = String::from("こんにちは");
let hello = String::from("안녕하세요");
let hello = String::from("你好");
let hello = String::from("Olá");
let hello = String::from("Здравствуйте");
let hello = String::from("Hola");
#}

Listing 8-14: Storing greetings in different languages in strings

All of these are valid String values.

A String can grow in size and its contents can change, just like the contents of a Vec<T>, if you push more data into it. In addition, you can conveniently use the + operator or the format! macro to concatenate String values.

We can grow a String by using the push_str method to append a string slice, as shown in Listing 8-15:

# #![allow(unused_variables)]
#fn main() {
let mut s = String::from("foo");
s.push_str("bar");
#}

Listing 8-15: Appending a string slice to a String using the push_str method

After these two lines, s will contain foobar. The push_str method takes a string slice because we don’t necessarily want to take ownership of the parameter. For example, the code in Listing 8-16 shows that it would be unfortunate if we weren’t able to use s2 after appending its contents to s1:

# #![allow(unused_variables)]
#fn main() {
let mut s1 = String::from("foo");
let s2 = "bar";
s1.push_str(s2);
println!("s2 is {}", s2);
#}

Listing 8-16: Using a string slice after appending its contents to a String

If the push_str method took ownership of s2, we wouldn’t be able to print its value on the last line. However, this code works as we’d expect!

The push method takes a single character as a parameter and adds it to the String. Listing 8-17 shows code that adds the letter l to a String using the push method:

# #![allow(unused_variables)]
#fn main() {
let mut s = String::from("lo");
s.push('l');
#}

Listing 8-17: Adding one character to a String value using push

As a result of this code, s will contain lol.

Often, you’ll want to combine two existing strings. One way is to use the + operator, as shown in Listing 8-18:

# #![allow(unused_variables)]
#fn main() {
let s1 = String::from("Hello, ");
let s2 = String::from("world!");
let s3 = s1 + &s2; // Note s1 has been moved here and can no longer be used
#}

Listing 8-18: Using the + operator to combine two String values into a new String value

The string s3 will contain Hello, world! as a result of this code. The reason s1 is no longer valid after the addition and the reason we used a reference to s2 has to do with the signature of the method that gets called when we use the + operator. The + operator uses the add method, whose signature looks something like this:

fn add(self, s: &str) -> String {

This isn’t the exact signature that’s in the standard library: in the standard library, add is defined using generics. Here, we’re looking at the signature of add with concrete types substituted for the generic ones, which is what happens when we call this method with String values. We’ll discuss generics in Chapter 10. This signature gives us the clues we need to understand the tricky bits of the + operator.

First, s2 has an &, meaning that we’re adding a reference of the second string to the first string because of the s parameter in the add function: we can only add a &str to a String; we can’t add two String values together. But wait—the type of &s2 is &String, not &str, as specified in the second parameter to add. So why does Listing 8-18 compile?

The reason we’re able to use &s2 in the call to add is that the compiler can coerce the &String argument into a &str. When we call the add method, Rust uses a deref coercion, which here turns &s2 into &s2[..]. We’ll discuss deref coercion in more depth in Chapter 15. Because add does not take ownership of the s parameter, s2 will still be a valid String after this operation.

Second, we can see in the signature that add takes ownership of self, because self does not have an &. This means s1 in Listing 8-18 will be moved into the add call and no longer be valid after that. So although let s3 = s1 + &s2; looks like it will copy both strings and create a new one, this statement actually takes ownership of s1, appends a copy of the contents of s2, and then returns ownership of the result. In other words, it looks like it’s making a lot of copies but isn’t; the implementation is more efficient than copying.

If we need to concatenate multiple strings, the behavior of the + operator gets unwieldy:

# #![allow(unused_variables)]
#fn main() {
let s1 = String::from("tic");
let s2 = String::from("tac");
let s3 = String::from("toe");

let s = s1 + "-" + &s2 + "-" + &s3;
#}

At this point, s will be tic-tac-toe. With all of the + and " characters, it’s difficult to see what’s going on. For more complicated string combining, we can use the format! macro:

# #![allow(unused_variables)]
#fn main() {
let s1 = String::from("tic");
let s2 = String::from("tac");
let s3 = String::from("toe");

let s = format!("{}-{}-{}", s1, s2, s3);
#}

This code also sets s to tic-tac-toe. The format! macro works in the same way as println!, but instead of printing the output to the screen, it returns a String with the contents. The version of the code using format! is much easier to read and doesn’t take ownership of any of its parameters.

In many other programming languages, accessing individual characters in a string by referencing them by index is a valid and common operation. However, if you try to access parts of a String using indexing syntax in Rust, you’ll get an error. Consider the invalid code in Listing 8-19:

let s1 = String::from("hello");
let h = s1[0];

Listing 8-19: Attempting to use indexing syntax with a String

This code will result in the following error:

error[E0277]: the trait bound `std::string::String: std::ops::Index<{integer}>` is not satisfied
 -->
  |
3 |     let h = s1[0];
  |             ^^^^^ the type `std::string::String` cannot be indexed by `{integer}`
  |
  = help: the trait `std::ops::Index<{integer}>` is not implemented for `std::string::String`

The error and the note tell the story: Rust strings don’t support indexing. But why not? To answer that question, we need to discuss how Rust stores strings in memory.

A String is a wrapper over a Vec<u8>. Let’s look at some of our properly encoded UTF-8 example strings from Listing 8-14. First, this one:

# #![allow(unused_variables)]
#fn main() {
let len = String::from("Hola").len();
#}

In this case, len will be 4, which means the vector storing the string “Hola” is 4 bytes long. Each of these letters takes 1 byte when encoded in UTF-8. But what about the following line? (Note that this line begins with the capital Cyrillic letter Ze, not the Arabic number 3.)

# #![allow(unused_variables)]
#fn main() {
let len = String::from("Здравствуйте").len();
#}

Asked how long the string is, you might say 12. However, Rust’s answer is 24: that’s the number of bytes it takes to encode “Здравствуйте” in UTF-8, because each Unicode scalar value takes 2 bytes of storage. Therefore, an index into the string’s bytes will not always correlate to a valid Unicode scalar value. To demonstrate, consider this invalid Rust code:

let hello = "Здравствуйте";
let answer = &hello[0];

What should the value of answer be? Should it be З, the first letter? When encoded in UTF-8, the first byte of З is 208 and the second is 151, so answer should in fact be 208, but 208 is not a valid character on its own. Returning 208 is likely not what a user would want if they asked for the first letter of this string; however, that’s the only data that Rust has at byte index 0. Users generally don’t want the byte value returned, even if the string contains only Latin letters: if &"hello"[0] were valid code that returned the byte value, it would return 104, not h. To avoid returning an unexpected value and causing bugs that might not be discovered immediately, Rust doesn’t compile this code at all and prevents misunderstandings early in the development process.

Another point about UTF-8 is that there are actually three relevant ways to look at strings from Rust’s perspective: as bytes, scalar values, and grapheme clusters (the closest thing to what we would call letters).

If we look at the Hindi word “नमस्ते” written in the Devanagari script, it is stored as a vector of u8 values that looks like this:

[224, 164, 168, 224, 164, 174, 224, 164, 184, 224, 165, 141, 224, 164, 164,
224, 165, 135]

That’s 18 bytes and is how computers ultimately store this data. If we look at them as Unicode scalar values, which are what Rust’s char type is, those bytes look like this:

['न', 'म', 'स', '्', 'त', 'े']

There are six char values here, but the fourth and sixth are not letters: they’re diacritics that don’t make sense on their own. Finally, if we look at them as grapheme clusters, we’d get what a person would call the four letters that make up the Hindi word:

["न", "म", "स्", "ते"]

Rust provides different ways of interpreting the raw string data that computers store so that each program can choose the interpretation it needs, no matter what human language the data is in.

A final reason Rust doesn’t allow us to index into a String to get a character is that indexing operations are expected to always take constant time (O(1)). But it isn’t possible to guarantee that performance with a String, because Rust would have to walk through the contents from the beginning to the index to determine how many valid characters there were.

Indexing into a string is often a bad idea because it’s not clear what the return type of the string-indexing operation should be: a byte value, a character, a grapheme cluster, or a string slice. Therefore, Rust asks you to be more specific if you really need to use indices to create string slices. To be more specific in your indexing and indicate that you want a string slice, rather than indexing using [] with a single number, you can use [] with a range to create a string slice containing particular bytes:

# #![allow(unused_variables)]
#fn main() {
let hello = "Здравствуйте";

let s = &hello[0..4];
#}

Here, s will be a &str that contains the first 4 bytes of the string. Earlier, we mentioned that each of these characters was 2 bytes, which means s will be Зд.

What would happen if we used &hello[0..1]? The answer: Rust would panic at runtime in the same way as if an invalid index were accessed in a vector:

thread 'main' panicked at 'byte index 1 is not a char boundary; it is inside 'З' (bytes 0..2) of `Здравствуйте`', src/libcore/str/mod.rs:2188:4

You should use ranges to create string slices with caution, because doing so can crash your program.

Fortunately, you can access elements in a string in other ways.

If you need to perform operations on individual Unicode scalar values, the best way to do so is to use the chars method. Calling chars on “नमस्ते” separates out and returns six values of type char, and you can iterate over the result in order to access each element:

# #![allow(unused_variables)]
#fn main() {
for c in "नमस्ते".chars() {
    println!("{}", c);
}
#}

This code will print the following:

न
म
स
्
त
े

The bytes method returns each raw byte, which might be appropriate for your domain:

# #![allow(unused_variables)]
#fn main() {
for b in "नमस्ते".bytes() {
    println!("{}", b);
}
#}

This code will print the 18 bytes that make up this String:

224
164
// --snip--
165
135

But be sure to remember that valid Unicode scalar values may be made up of more than 1 byte.

Getting grapheme clusters from strings is complex, so this functionality is not provided by the standard library. Crates are available on crates.io if this is the functionality you need.

To summarize, strings are complicated. Different programming languages make different choices about how to present this complexity to the programmer. Rust has chosen to make the correct handling of String data the default behavior for all Rust programs, which means programmers have to put more thought into handling UTF-8 data upfront. This trade-off exposes more of the complexity of strings than is apparent in other programming languages, but it prevents you from having to handle errors involving non-ASCII characters later in your development life cycle.

Let’s switch to something a bit less complex: hash maps!

The last of our common collections is the hash map. The type HashMap<K, V> stores a mapping of keys of type K to values of type V. It does this via a hashing function, which determines how it places these keys and values into memory. Many programming languages support this kind of data structure, but they often use a different name, such as hash, map, object, hash table, or associative array, just to name a few.

Hash maps are useful when you want to look up data not by using an index, as you can with vectors, but by using a key that can be of any type. For example, in a game, you could keep track of each team’s score in a hash map in which each key is a team’s name and the values are each team’s score. Given a team name, you can retrieve its score.

We’ll go over the basic API of hash maps in this section, but many more goodies are hiding in the functions defined on HashMap<K, V> by the standard library. As always, check the standard library documentation for more information.

You can create an empty hash map with new and add elements with insert. In Listing 8-20, we’re keeping track of the scores of two teams whose names are Blue and Yellow. The Blue team starts with 10 points, and the Yellow team starts with 50:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let mut scores = HashMap::new();

scores.insert(String::from("Blue"), 10);
scores.insert(String::from("Yellow"), 50);
#}

Listing 8-20: Creating a new hash map and inserting some keys and values

Note that we need to first use the HashMap from the collections portion of the standard library. Of our three common collections, this one is the least often used, so it’s not included in the features brought into scope automatically in the prelude. Hash maps also have less support from the standard library; there’s no built-in macro to construct them, for example.

Just like vectors, hash maps store their data on the heap. This HashMap has keys of type String and values of type i32. Like vectors, hash maps are homogeneous: all of the keys must have the same type, and all of the values must have the same type.

Another way of constructing a hash map is by using the collect method on a vector of tuples, where each tuple consists of a key and its value. The collect method gathers data into a number of collection types, including HashMap. For example, if we had the team names and initial scores in two separate vectors, we could use the zip method to create a vector of tuples where “Blue” is paired with 10, and so forth. Then we could use the collect method to turn that vector of tuples into a hash map, as shown in Listing 8-21:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let teams  = vec![String::from("Blue"), String::from("Yellow")];
let initial_scores = vec![10, 50];

let scores: HashMap<_, _> = teams.iter().zip(initial_scores.iter()).collect();
#}

Listing 8-21: Creating a hash map from a list of teams and a list of scores

The type annotation HashMap<_, _> is needed here because it’s possible to collect into many different data structures and Rust doesn’t know which you want unless you specify. For the parameters for the key and value types, however, we use underscores, and Rust can infer the types that the hash map contains based on the types of the data in the vectors.

For types that implement the Copy trait, like i32, the values are copied into the hash map. For owned values like String, the values will be moved and the hash map will be the owner of those values, as demonstrated in Listing 8-22:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let field_name = String::from("Favorite color");
let field_value = String::from("Blue");

let mut map = HashMap::new();
map.insert(field_name, field_value);
// field_name and field_value are invalid at this point, try using them and
// see what compiler error you get!
#}

Listing 8-22: Showing that keys and values are owned by the hash map once they’re inserted

We aren’t able to use the variables field_name and field_value after they’ve been moved into the hash map with the call to insert.

If we insert references to values into the hash map, the values won’t be moved into the hash map. The values that the references point to must be valid for at least as long as the hash map is valid. We’ll talk more about these issues in the “Validating References with Lifetimes” section in Chapter 10.

We can get a value out of the hash map by providing its key to the get method, as shown in Listing 8-23:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let mut scores = HashMap::new();

scores.insert(String::from("Blue"), 10);
scores.insert(String::from("Yellow"), 50);

let team_name = String::from("Blue");
let score = scores.get(&team_name);
#}

Listing 8-23: Accessing the score for the Blue team stored in the hash map

Here, score will have the value that’s associated with the Blue team, and the result will be Some(&10). The result is wrapped in Some because get returns an Option<&V>; if there’s no value for that key in the hash map, get will return None. The program will need to handle the Option in one of the ways that we covered in Chapter 6.

We can iterate over each key/value pair in a hash map in a similar manner as we do with vectors, using a for loop:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let mut scores = HashMap::new();

scores.insert(String::from("Blue"), 10);
scores.insert(String::from("Yellow"), 50);

for (key, value) in &scores {
    println!("{}: {}", key, value);
}
#}

This code will print each pair in an arbitrary order:

Yellow: 50
Blue: 10

Although the number of keys and values is growable, each key can only have one value associated with it at a time. When you want to change the data in a hash map, you have to decide how to handle the case when a key already has a value assigned. You could replace the old value with the new value, completely disregarding the old value. You could keep the old value and ignore the new value, only adding the new value if the key doesn’t already have a value. Or you could combine the old value and the new value. Let’s look at how to do each of these!

If we insert a key and a value into a hash map and then insert that same key with a different value, the value associated with that key will be replaced. Even though the code in Listing 8-24 calls insert twice, the hash map will only contain one key/value pair because we’re inserting the value for the Blue team’s key both times:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let mut scores = HashMap::new();

scores.insert(String::from("Blue"), 10);
scores.insert(String::from("Blue"), 25);

println!("{:?}", scores);
#}

Listing 8-24: Replacing a value stored with a particular key

This code will print {"Blue": 25}. The original value of 10 has been overwritten.

It’s common to check whether a particular key has a value and, if it doesn’t, insert a value for it. Hash maps have a special API for this called entry that takes the key you want to check as a parameter. The return value of the entry function is an enum called Entry that represents a value that might or might not exist. Let’s say we want to check whether the key for the Yellow team has a value associated with it. If it doesn’t, we want to insert the value 50, and the same for the Blue team. Using the entry API, the code looks like Listing 8-25:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let mut scores = HashMap::new();
scores.insert(String::from("Blue"), 10);

scores.entry(String::from("Yellow")).or_insert(50);
scores.entry(String::from("Blue")).or_insert(50);

println!("{:?}", scores);
#}

Listing 8-25: Using the entry method to only insert if the key does not already have a value

The or_insert method on Entry is defined to return a mutable reference to the value for the corresponding Entry key if that key exists, and if not, inserts the parameter as the new value for this key and returns a mutable reference to the new value. This technique is much cleaner than writing the logic ourselves and, in addition, plays more nicely with the borrow checker.

Running the code in Listing 8-25 will print {"Yellow": 50, "Blue": 10}. The first call to entry will insert the key for the Yellow team with the value 50 because the Yellow team doesn’t have a value already. The second call to entry will not change the hash map because the Blue team already has the value 10.

Another common use case for hash maps is to look up a key’s value and then update it based on the old value. For instance, Listing 8-26 shows code that counts how many times each word appears in some text. We use a hash map with the words as keys and increment the value to keep track of how many times we’ve seen that word. If it’s the first time we’ve seen a word, we’ll first insert the value 0:

# #![allow(unused_variables)]
#fn main() {
use std::collections::HashMap;

let text = "hello world wonderful world";

let mut map = HashMap::new();

for word in text.split_whitespace() {
    let count = map.entry(word).or_insert(0);
    *count += 1;
}

println!("{:?}", map);
#}

Listing 8-26: Counting occurrences of words using a hash map that stores words and counts

This code will print {"world": 2, "hello": 1, "wonderful": 1}. The or_insert method actually returns a mutable reference (&mut V) to the value for this key. Here we store that mutable reference in the count variable, so in order to assign to that value, we must first dereference count using the asterisk (*). The mutable reference goes out of scope at the end of the for loop, so all of these changes are safe and allowed by the borrowing rules.

By default, HashMap uses a cryptographically secure hashing function that can provide resistance to Denial of Service (DoS) attacks. This is not the fastest hashing algorithm available, but the trade-off for better security that comes with the drop in performance is worth it. If you profile your code and find that the default hash function is too slow for your purposes, you can switch to another function by specifying a different hasher. A hasher is a type that implements the BuildHasher trait. We’ll talk about traits and how to implement them in Chapter 10. You don’t necessarily have to implement your own hasher from scratch; crates.io has libraries shared by other Rust users that provide hashers implementing many common hashing algorithms.

Vectors, strings, and hash maps will provide a large amount of functionality necessary in programs when you need to store, access, and modify data. Here are some exercises you should now be equipped to solve:

• Given a list of integers, use a vector and return the mean (the average value), median (when sorted, the value in the middle position), and mode (the value that occurs most often; a hash map will be helpful here) of the list.
• Convert strings to pig latin. The first consonant of each word is moved to the end of the word and “ay” is added, so “first” becomes “irst-fay.” Words that start with a vowel have “hay” added to the end instead (“apple” becomes “apple-hay”). Keep in mind the details about UTF-8 encoding!
• Using a hash map and vectors, create a text interface to allow a user to add employee names to a department in a company. For example, “Add Sally to Engineering” or “Add Amir to Sales.” Then let the user retrieve a list of all people in a department or all people in the company by department, sorted alphabetically.

The standard library API documentation describes methods that vectors, strings, and hash maps have that will be helpful for these exercises!

We’re getting into more complex programs in which operations can fail, so, it’s a perfect time to discuss error handling. We’ll do that next!

Rust’s commitment to reliability extends to error handling. Errors are a fact of life in software, so Rust has a number of features for handling situations in which something goes wrong. In many cases, Rust requires you to acknowledge the possibility of an error and take some action before your code will compile. This requirement makes your program more robust by ensuring that you’ll discover errors and handle them appropriately before you’ve deployed your code to production!

Rust groups errors into two major categories: recoverable and unrecoverable errors. For a recoverable error, such as a file not found error, it’s reasonable to report the problem to the user and retry the operation. Unrecoverable errors are always symptoms of bugs, like trying to access a location beyond the end of an array.

Most languages don’t distinguish between these two kinds of errors and handle both in the same way, using mechanisms such as exceptions. Rust doesn’t have exceptions. Instead, it has the type Result<T, E> for recoverable errors and the panic! macro that stops execution when the program encounters an unrecoverable error. This chapter covers calling panic! first and then talks about returning Result<T, E> values. Additionally, we’ll explore considerations when deciding whether to try to recover from an error or to stop execution.

Sometimes, bad things happen in your code, and there’s nothing you can do about it. In these cases, Rust has the panic! macro. When the panic! macro executes, your program will print a failure message, unwind and clean up the stack, and then quit. This most commonly occurs when a bug of some kind has been detected and it’s not clear to the programmer how to handle the error.

Unwinding the Stack or Aborting in Response to a Panic

By default, when a panic occurs, the program starts unwinding, which means Rust walks back up the stack and cleans up the data from each function it encounters. But this walking back and cleanup is a lot of work. The alternative is to immediately abort, which ends the program without cleaning up. Memory that the program was using will then need to be cleaned up by the operating system. If in your project you need to make the resulting binary as small as possible, you can switch from unwinding to aborting upon a panic by adding panic = 'abort' to the appropriate [profile] sections in your Cargo.toml file. For example, if you want to abort on panic in release mode, add this:

[profile.release]
panic = 'abort'

Let’s try calling panic! in a simple program:

Filename: src/main.rs

fn main() {
    panic!("crash and burn");
}

When you run the program, you’ll see something like this:

$ cargo run
   Compiling panic v0.1.0 (file:///projects/panic)
    Finished dev [unoptimized + debuginfo] target(s) in 0.25 secs
     Running `target/debug/panic`
thread 'main' panicked at 'crash and burn', src/main.rs:2:4
note: Run with `RUST_BACKTRACE=1` for a backtrace.

The call to panic! causes the error message contained in the last three lines. The first line shows our panic message and the place in our source code where the panic occurred: src/main.rs:2:4 indicates that it’s the second line, fourth character of our src/main.rs file.

In this case, the line indicated is part of our code, and if we go to that line, we see the panic! macro call. In other cases, the panic! call might be in code that our code calls, and the filename and line number reported by the error message will be someone else’s code where the panic! macro is called, not the line of our code that eventually led to the panic! call. We can use the backtrace of the functions the panic! call came from to figure out the part of our code that is causing the problem. We’ll discuss what a backtrace is in more detail next.

Let’s look at another example to see what it’s like when a panic! call comes from a library because of a bug in our code instead of from our code calling the macro directly. Listing 9-1 has some code that attempts to access an element by index in a vector:

Filename: src/main.rs

fn main() {
    let v = vec![1, 2, 3];

    v[99];
}

Listing 9-1: Attempting to access an element beyond the end of a vector, which will cause a panic!

Here, we’re attempting to access the hundredth element of our vector (which is at index 99 because indexing starts at zero), but it has only three elements. In this situation, Rust will panic. Using [] is supposed to return an element, but if you pass an invalid index, there’s no element that Rust could return here that would be correct.

Other languages, like C, will attempt to give you exactly what you asked for in this situation, even though it isn’t what you want: you’ll get whatever is at the location in memory that would correspond to that element in the vector, even though the memory doesn’t belong to the vector. This is called a buffer overread and can lead to security vulnerabilities if an attacker is able to manipulate the index in such a way as to read data they shouldn’t be allowed to that is stored after the array.

To protect your program from this sort of vulnerability, if you try to read an element at an index that doesn’t exist, Rust will stop execution and refuse to continue. Let’s try it and see:

$ cargo run
   Compiling panic v0.1.0 (file:///projects/panic)
    Finished dev [unoptimized + debuginfo] target(s) in 0.27 secs
     Running `target/debug/panic`
thread 'main' panicked at 'index out of bounds: the len is 3 but the index is
99', /checkout/src/liballoc/vec.rs:1555:10
note: Run with `RUST_BACKTRACE=1` for a backtrace.

This error points at a file we didn’t write, vec.rs. That’s the implementation of Vec<T> in the standard library. The code that gets run when we use [] on our vector v is in vec.rs, and that is where the panic! is actually happening.

The next note line tells us that we can set the RUST_BACKTRACE environment variable to get a backtrace of exactly what happened to cause the error. A backtrace is a list of all the functions that have been called to get to this point. Backtraces in Rust work as they do in other languages: the key to reading the backtrace is to start from the top and read until you see files you wrote. That’s the spot where the problem originated. The lines above the lines mentioning your files are code that your code called; the lines below are code that called your code. These lines might include core Rust code, standard library code, or crates that you’re using. Let’s try getting a backtrace by setting the RUST_BACKTRACE environment variable to any value except 0. Listing 9-2 shows output similar to what you’ll see:

$ RUST_BACKTRACE=1 cargo run
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/panic`
thread 'main' panicked at 'index out of bounds: the len is 3 but the index is 99', /checkout/src/liballoc/vec.rs:1555:10
stack backtrace:
   0: std::sys::imp::backtrace::tracing::imp::unwind_backtrace
             at /checkout/src/libstd/sys/unix/backtrace/tracing/gcc_s.rs:49
   1: std::sys_common::backtrace::_print
             at /checkout/src/libstd/sys_common/backtrace.rs:71
   2: std::panicking::default_hook::{{closure}}
             at /checkout/src/libstd/sys_common/backtrace.rs:60
             at /checkout/src/libstd/panicking.rs:381
   3: std::panicking::default_hook
             at /checkout/src/libstd/panicking.rs:397
   4: std::panicking::rust_panic_with_hook
             at /checkout/src/libstd/panicking.rs:611
   5: std::panicking::begin_panic
             at /checkout/src/libstd/panicking.rs:572
   6: std::panicking::begin_panic_fmt
             at /checkout/src/libstd/panicking.rs:522
   7: rust_begin_unwind
             at /checkout/src/libstd/panicking.rs:498
   8: core::panicking::panic_fmt
             at /checkout/src/libcore/panicking.rs:71
   9: core::panicking::panic_bounds_check
             at /checkout/src/libcore/panicking.rs:58
  10: <alloc::vec::Vec<T> as core::ops::index::Index<usize>>::index
             at /checkout/src/liballoc/vec.rs:1555
  11: panic::main
             at src/main.rs:4
  12: __rust_maybe_catch_panic
             at /checkout/src/libpanic_unwind/lib.rs:99
  13: std::rt::lang_start
             at /checkout/src/libstd/panicking.rs:459
             at /checkout/src/libstd/panic.rs:361
             at /checkout/src/libstd/rt.rs:61
  14: main
  15: __libc_start_main
  16: <unknown>

Listing 9-2: The backtrace generated by a call to panic! displayed when the environment variable RUST_BACKTRACE is set

That’s a lot of output! The exact output you see might be different depending on your operating system and Rust version. In order to get backtraces with this information, debug symbols must be enabled. Debug symbols are enabled by default when using cargo build or cargo run without the --release flag, as we have here.

In the output in Listing 9-2, line 11 of the backtrace points to the line in our project that’s causing the problem: line 4 of src/main.rs. If we don’t want our program to panic, the location pointed to by the first line mentioning a file we wrote is where we should start investigating. In Listing 9-1, where we deliberately wrote code that would panic in order to demonstrate how to use backtraces, the way to fix the panic is to not request an element at index 99 from a vector that only contains 3 items. When your code panics in the future, you’ll need to figure out what action the code is taking with what values to cause the panic and what the code should do instead.

We’ll come back to panic! and when we should and should not use panic! to handle error conditions in the “To panic! or Not to panic!” section later in this chapter. Next, we’ll look at how to recover from an error using Result.

Most errors aren’t serious enough to require the program to stop entirely. Sometimes, when a function fails, it’s for a reason that you can easily interpret and respond to. For example, if you try to open a file and that operation fails because the file doesn’t exist, you might want to create the file instead of terminating the process.

Recall from “Handling Potential Failure with the Result Type” in Chapter 2 that the Result enum is defined as having two variants, Ok and Err, as follows:

# #![allow(unused_variables)]
#fn main() {
enum Result<T, E> {
    Ok(T),
    Err(E),
}
#}

The T and E are generic type parameters: we’ll discuss generics in more detail in Chapter 10. What you need to know right now is that T represents the type of the value that will be returned in a success case within the Ok variant, and E represents the type of the error that will be returned in a failure case within the Err variant. Because Result has these generic type parameters, we can use the Result type and the functions that the standard library has defined on it in many different situations where the successful value and error value we want to return may differ.

Let’s call a function that returns a Result value because the function could fail. In Listing 9-3 we try to open a file:

Filename: src/main.rs

use std::fs::File;

fn main() {
    let f = File::open("hello.txt");
}

Listing 9-3: Opening a file

How do we know File::open returns a Result? We could look at the standard library API documentation, or we could ask the compiler! If we give f a type annotation that we know is not the return type of the function and then try to compile the code, the compiler will tell us that the types don’t match. The error message will then tell us what the type of f is. Let’s try it! We know that the return type of File::open isn’t of type u32, so let’s change the let f statement to this:

let f: u32 = File::open("hello.txt");

Attempting to compile now gives us the following output:

error[E0308]: mismatched types
 --> src/main.rs:4:18
  |
4 |     let f: u32 = File::open("hello.txt");
  |                  ^^^^^^^^^^^^^^^^^^^^^^^ expected u32, found enum
`std::result::Result`
  |
  = note: expected type `u32`
             found type `std::result::Result<std::fs::File, std::io::Error>`

This tells us the return type of the File::open function is a Result<T, E>. The generic parameter T has been filled in here with the type of the success value, std::fs::File, which is a file handle. The type of E used in the error value is std::io::Error.

This return type means the call to File::open might succeed and return a file handle that we can read from or write to. The function call also might fail: for example, the file might not exist, or we might not have permission to access the file. The File::open function needs to have a way to tell us whether it succeeded or failed and at the same time give us either the file handle or error information. This information is exactly what the Result enum conveys.

In the case where File::open succeeds, the value in the variable f will be an instance of Ok that contains a file handle. In the case where it fails, the value in f will be an instance of Err that contains more information about the kind of error that happened.

We need to add to the code in Listing 9-3 to take different actions depending on the value File::open returns. Listing 9-4 shows one way to handle the Result using a basic tool, the match expression that we discussed in Chapter 6.

Filename: src/main.rs

use std::fs::File;

fn main() {
    let f = File::open("hello.txt");

    let f = match f {
        Ok(file) => file,
        Err(error) => {
            panic!("There was a problem opening the file: {:?}", error)
        },
    };
}

Listing 9-4: Using a match expression to handle the Result variants that might be returned

Note that, like the Option enum, the Result enum and its variants have been imported in the prelude, so we don’t need to specify Result:: before the Ok and Err variants in the match arms.

Here we tell Rust that when the result is Ok, return the inner file value out of the Ok variant, and we then assign that file handle value to the variable f. After the match, we can use the file handle for reading or writing.

The other arm of the match handles the case where we get an Err value from File::open. In this example, we’ve chosen to call the panic! macro. If there’s no file named hello.txt in our current directory and we run this code, we’ll see the following output from the panic! macro:

thread 'main' panicked at 'There was a problem opening the file: Error { repr:
Os { code: 2, message: "No such file or directory" } }', src/main.rs:9:12

As usual, this output tells us exactly what has gone wrong.

The code in Listing 9-4 will panic! no matter why File::open failed. What we want to do instead is take different actions for different failure reasons: if File::open failed because the file doesn’t exist, we want to create the file and return the handle to the new file. If File::open failed for any other reason—for example, because we didn’t have permission to open the file—we still want the code to panic! in the same way as it did in Listing 9-4. Look at Listing 9-5, which adds another arm to the match:

Filename: src/main.rs

use std::fs::File;
use std::io::ErrorKind;

fn main() {
    let f = File::open("hello.txt");

    let f = match f {
        Ok(file) => file,
        Err(ref error) if error.kind() == ErrorKind::NotFound => {
            match File::create("hello.txt") {
                Ok(fc) => fc,
                Err(e) => {
                    panic!(
                        "Tried to create file but there was a problem: {:?}",
                        e
                    )
                },
            }
        },
        Err(error) => {
            panic!(
                "There was a problem opening the file: {:?}",
                error
            )
        },
    };
}

Listing 9-5: Handling different kinds of errors in different ways

The type of the value that File::open returns inside the Err variant is io::Error, which is a struct provided by the standard library. This struct has a method kind that we can call to get an io::ErrorKind value. The enum io::ErrorKind is provided by the standard library and has variants representing the different kinds of errors that might result from an io operation. The variant we want to use is ErrorKind::NotFound, which indicates the file we’re trying to open doesn’t exist yet.

The condition if error.kind() == ErrorKind::NotFound is called a match guard: it’s an extra condition on a match arm that further refines the arm’s pattern. This condition must be true for that arm’s code to be run; otherwise, the pattern matching will move on to consider the next arm in the match. The ref in the pattern is needed so error is not moved into the guard condition but is merely referenced by it. The reason you use ref to create a reference in a pattern instead of & will be covered in detail in Chapter 18. In short, in the context of a pattern, & matches a reference and gives you its value, but ref matches a value and gives you a reference to it.

The condition we want to check in the match guard is whether the value returned by error.kind() is the NotFound variant of the ErrorKind enum. If it is, we try to create the file with File::create. However, because File::create could also fail, we need to add an inner match statement as well. When the file can’t be opened, a different error message will be printed. The last arm of the outer match stays the same so the program panics on any error besides the missing file error.

Using match works well enough, but it can be a bit verbose and doesn’t always communicate intent well. The Result<T, E> type has many helper methods defined on it to do various tasks. One of those methods, called unwrap, is a shortcut method that is implemented just like the match statement we wrote in Listing 9-4. If the Result value is the Ok variant, unwrap will return the value inside the Ok. If the Result is the Err variant, unwrap will call the panic! macro for us. Here is an example of unwrap in action:

Filename: src/main.rs

use std::fs::File;

fn main() {
    let f = File::open("hello.txt").unwrap();
}

If we run this code without a hello.txt file, we’ll see an error message from the panic! call that the unwrap method makes:

thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Error {
repr: Os { code: 2, message: "No such file or directory" } }',
src/libcore/result.rs:906:4

Another method, expect, which is similar to unwrap, lets us also choose the panic! error message. Using expect instead of unwrap and providing good error messages can convey your intent and make tracking down the source of a panic easier. The syntax of expect looks like this:

Filename: src/main.rs

use std::fs::File;

fn main() {
    let f = File::open("hello.txt").expect("Failed to open hello.txt");
}

We use expect in the same way as unwrap: to return the file handle or call the panic! macro. The error message used by expect in its call to panic! will be the parameter that we pass to expect, rather than the default panic! message that unwrap uses. Here’s what it looks like:

thread 'main' panicked at 'Failed to open hello.txt: Error { repr: Os { code:
2, message: "No such file or directory" } }', src/libcore/result.rs:906:4

Because this error message starts with the text we specified, Failed to open hello.txt, it will be easier to find where in the code this error message is coming from. If we use unwrap in multiple places, it can take more time to figure out exactly which unwrap is causing the panic because all unwrap calls that panic print the same message.

When you’re writing a function whose implementation calls something that might fail, instead of handling the error within this function, you can return the error to the calling code so that it can decide what to do. This is known as propagating the error and gives more control to the calling code, where there might be more information or logic that dictates how the error should be handled than what you have available in the context of your code.

For example, Listing 9-6 shows a function that reads a username from a file. If the file doesn’t exist or can’t be read, this function will return those errors to the code that called this function:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
use std::io;
use std::io::Read;
use std::fs::File;

fn read_username_from_file() -> Result<String, io::Error> {
    let f = File::open("hello.txt");

    let mut f = match f {
        Ok(file) => file,
        Err(e) => return Err(e),
    };

    let mut s = String::new();

    match f.read_to_string(&mut s) {
        Ok(_) => Ok(s),
        Err(e) => Err(e),
    }
}
#}

Listing 9-6: A function that returns errors to the calling code using match

Let’s look at the return type of the function first: Result<String, io::Error>. This means the function is returning a value of the type Result<T, E> where the generic parameter T has been filled in with the concrete type String, and the generic type E has been filled in with the concrete type io::Error. If this function succeeds without any problems, the code that calls this function will receive an Ok value that holds a String—the username that this function read from the file. If this function encounters any problems, the code that calls this function will receive an Err value that holds an instance of io::Error that contains more information about what the problems were. We chose io::Error as the return type of this function because that happens to be the type of the error value returned from both of the operations we’re calling in this function’s body that might fail: the File::open function and the read_to_string method.

The body of the function starts by calling the File::open function. Then we handle the Result value returned with a match similar to the match in Listing 9-4, only instead of calling panic! in the Err case, we return early from this function and pass the error value from File::open back to the calling code as this function’s error value. If File::open succeeds, we store the file handle in the variable f and continue.

Then we create a new String in variable s and call the read_to_string method on the file handle in f to read the contents of the file into s. The read_to_string method also returns a Result because it might fail, even though File::open succeeded. So we need another match to handle that Result: if read_to_string succeeds, then our function has succeeded, and we return the username from the file that’s now in s wrapped in an Ok. If read_to_string fails, we return the error value in the same way that we returned the error value in the match that handled the return value of File::open. However, we don’t need to explicitly say return, because this is the last expression in the function.

The code that calls this code will then handle getting either an Ok value that contains a username or an Err value that contains an io::Error. We don’t know what the calling code will do with those values. If the calling code gets an Err value, it could call panic! and crash the program, use a default username, or look up the username from somewhere other than a file, for example. We don’t have enough information on what the calling code is actually trying to do, so we propagate all the success or error information upward for it to handle appropriately.

This pattern of propagating errors is so common in Rust that Rust provides the question mark operator ? to make this easier.

Listing 9-7 shows an implementation of read_username_from_file that has the same functionality as it had in Listing 9-6, but this implementation uses the question mark operator:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
use std::io;
use std::io::Read;
use std::fs::File;

fn read_username_from_file() -> Result<String, io::Error> {
    let mut f = File::open("hello.txt")?;
    let mut s = String::new();
    f.read_to_string(&mut s)?;
    Ok(s)
}
#}

Listing 9-7: A function that returns errors to the calling code using ?

The ? placed after a Result value is defined to work in almost the same way as the match expressions we defined to handle the Result values in Listing 9-6. If the value of the Result is an Ok, the value inside the Ok will get returned from this expression, and the program will continue. If the value is an Err, the value inside the Err will be returned from the whole function as if we had used the return keyword so the error value gets propagated to the calling code.

There is a difference between what the match expression from Listing 9-6 and ? do: error values used with ? go through the from function, defined in the From trait in the standard library, which is used to convert errors from one type into another. When ? calls the from function, the error type received is converted into the error type defined in the return type of the current function. This is useful when a function returns one error type to represent all the ways a function might fail, even if parts might fail for many different reasons. As long as each error type implements the from function to define how to convert itself to the returned error type, ? takes care of the conversion automatically.

In the context of Listing 9-7, the ? at the end of the File::open call will return the value inside an Ok to the variable f. If an error occurs, ? will return early out of the whole function and give any Err value to the calling code. The same thing applies to the ? at the end of the read_to_string call.

The ? operator eliminates a lot of boilerplate and makes this function’s implementation simpler. We could even shorten this code further by chaining method calls immediately after the ?, as shown in Listing 9-8:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
use std::io;
use std::io::Read;
use std::fs::File;

fn read_username_from_file() -> Result<String, io::Error> {
    let mut s = String::new();

    File::open("hello.txt")?.read_to_string(&mut s)?;

    Ok(s)
}
#}

Listing 9-8: Chaining method calls after ?

We’ve moved the creation of the new String in s to the beginning of the function; that part hasn’t changed. Instead of creating a variable f, we’ve chained the call to read_to_string directly onto the result of File::open("hello.txt")?. We still have a ? at the end of the read_to_string call, and we still return an Ok value containing the username in s when both File::open and read_to_string succeed rather than returning errors. The functionality is again the same as in Listing 9-6 and Listing 9-7; this is just a different, more ergonomic way to write it.

The ? operator can only be used in functions that have a return type of Result, because it is defined to work in the same way as the match expression we defined in Listing 9-6. The part of the match that requires a return type of Result is return Err(e), so the return type of the function must be a Result to be compatible with this return.

Let’s look at what happens if we use ? in the main function, which you’ll recall has a return type of ():

use std::fs::File;

fn main() {
    let f = File::open("hello.txt")?;
}

When we compile this code, we get the following error message:

error[E0277]: the trait bound `(): std::ops::Try` is not satisfied
 --> src/main.rs:4:13
  |
4 |     let f = File::open("hello.txt")?;
  |             ------------------------
  |             |
  |             the `?` operator can only be used in a function that returns
  `Result` (or another type that implements `std::ops::Try`)
  |             in this macro invocation
  |
  = help: the trait `std::ops::Try` is not implemented for `()`
  = note: required by `std::ops::Try::from_error`

This error points out that we’re only allowed to use ? in a function that returns Result. In functions that don’t return Result, when you call other functions that return Result, you’ll need to use a match or one of the Result methods to handle the Result instead of using ? to potentially propagate the error to the calling code.

Now that we’ve discussed the details of calling panic! or returning Result, let’s return to the topic of how to decide which is appropriate to use in which cases.

So how do you decide when you should call panic! and when you should return Result? When code panics, there’s no way to recover. You could call panic! for any error situation, whether there’s a possible way to recover or not, but then you’re making the decision on behalf of the code calling your code that a situation is unrecoverable. When you choose to return a Result value, you give the calling code options rather than making the decision for it. The calling code could choose to attempt to recover in a way that’s appropriate for its situation, or it could decide that an Err value in this case is unrecoverable, so it can call panic! and turn your recoverable error into an unrecoverable one. Therefore, returning Result is a good default choice when you’re defining a function that might fail.

In rare situations, it’s more appropriate to write code that panics instead of returning a Result. Let’s explore why it’s appropriate to panic in examples, prototype code, and tests. Then we’ll discuss situations in which the compiler can’t tell that failure is impossible, but you as a human can. The chapter will conclude with some general guidelines on how to decide whether to panic in library code.

When you’re writing an example to illustrate some concept, having robust error-handling code in the example as well can make the example less clear. In examples, it’s understood that a call to a method like unwrap that could panic is meant as a placeholder for the way you’d want your application to handle errors, which can differ based on what the rest of your code is doing.

Similarly, the unwrap and expect methods are very handy when prototyping, before you’re ready to decide how to handle errors. They leave clear markers in your code for when you’re ready to make your program more robust.

If a method call fails in a test, you’d want the whole test to fail, even if that method isn’t the functionality under test. Because panic! is how a test is marked as a failure, calling unwrap or expect is exactly what should happen.

It would also be appropriate to call unwrap when you have some other logic that ensures the Result will have an Ok value, but the logic isn’t something the compiler understands. You’ll still have a Result value that you need to handle: whatever operation you’re calling still has the possibility of failing in general, even though it’s logically impossible in your particular situation. If you can ensure by manually inspecting the code that you’ll never have an Err variant, it’s perfectly acceptable to call unwrap. Here’s an example:

# #![allow(unused_variables)]
#fn main() {
use std::net::IpAddr;

let home: IpAddr = "127.0.0.1".parse().unwrap();
#}

We’re creating an IpAddr instance by parsing a hardcoded string. We can see that 127.0.0.1 is a valid IP address, so it’s acceptable to use unwrap here. However, having a hardcoded, valid string doesn’t change the return type of the parse method: we still get a Result value, and the compiler will still make us handle the Result as if the Err variant is a possibility because the compiler isn’t smart enough to see that this string is always a valid IP address. If the IP address string came from a user rather than being hardcoded into the program and therefore did have a possibility of failure, we’d definitely want to handle the Result in a more robust way instead.

It’s advisable to have your code panic when it’s possible that your code could end up in a bad state. In this context, a bad state is when some assumption, guarantee, contract, or invariant has been broken, such as when invalid values, contradictory values, or missing values are passed to your code—plus one or more of the following:

• The bad state is not something that’s expected to happen occasionally.
• Your code after this point needs to rely on not being in this bad state.
• There’s not a good way to encode this information in the types you use.

If someone calls your code and passes in values that don’t make sense, the best choice might be to call panic! and alert the person using your library to the bug in their code so they can fix it during development. Similarly, panic! is often appropriate if you’re calling external code that is out of your control and it returns an invalid state that you have no way of fixing.

When a bad state is reached, but it’s expected to happen no matter how well you write your code, it’s still more appropriate to return a Result rather than to make a panic! call. Examples include a parser being given malformed data or an HTTP request returning a status that indicates you have hit a rate limit. In these cases, you should indicate that failure is an expected possibility by returning a Result to propagate these bad states upward so the calling code can decide how to handle the problem. To call panic! wouldn’t be the best way to handle these cases.

When your code performs operations on values, your code should verify the values are valid first and panic if the values aren’t valid. This is mostly for safety reasons: attempting to operate on invalid data can expose your code to vulnerabilities. This is the main reason the standard library will call panic! if you attempt an out-of-bounds memory access: trying to access memory that doesn’t belong to the current data structure is a common security problem. Functions often have contracts: their behavior is only guaranteed if the inputs meet particular requirements. Panicking when the contract is violated makes sense because a contract violation always indicates a caller-side bug and it’s not a kind of error you want the calling code to have to explicitly handle. In fact, there’s no reasonable way for calling code to recover; the calling programmers need to fix the code. Contracts for a function, especially when a violation will cause a panic, should be explained in the API documentation for the function.

However, having lots of error checks in all of your functions would be verbose and annoying. Fortunately, you can use Rust’s type system (and thus the type checking the compiler does) to do many of the checks for you. If your function has a particular type as a parameter, you can proceed with your code’s logic knowing that the compiler has already ensured you have a valid value. For example, if you have a type rather than an Option, your program expects to have something rather than nothing. Your code then doesn’t have to handle two cases for the Some and None variants: it will only have one case for definitely having a value. Code trying to pass nothing to your function won’t even compile, so your function doesn’t have to check for that case at runtime. Another example is using an unsigned integer type such as u32, which ensures the parameter is never negative.

Let’s take the idea of using Rust’s type system to ensure we have a valid value one step further and look at creating a custom type for validation. Recall the guessing game in Chapter 2 in which our code asked the user to guess a number between 1 and 100. We never validated that the user’s guess was between those numbers before checking it against our secret number; we only validated that the guess was positive. In this case, the consequences were not very dire: our output of “Too high” or “Too low” would still be correct. But it would be a useful enhancement to guide the user toward valid guesses and have different behavior when a user guesses a number that’s out of range versus when a user types, for example, letters instead.

One way to do this would be to parse the guess as an i32 instead of only a u32 to allow potentially negative numbers, and then add a check for the number being in range, like so:

loop {
    // --snip--

    let guess: i32 = match guess.trim().parse() {
        Ok(num) => num,
        Err(_) => continue,
    };

    if guess < 1 || guess > 100 {
        println!("The secret number will be between 1 and 100.");
        continue;
    }

    match guess.cmp(&secret_number) {
    // --snip--
}

The if expression checks whether our value is out of range, tells the user about the problem, and calls continue to start the next iteration of the loop and ask for another guess. After the if expression, we can proceed with the comparisons between guess and the secret number knowing that guess is between 1 and 100.

However, this is not an ideal solution: if it was absolutely critical that the program only operated on values between 1 and 100, and it had many functions with this requirement, having a check like this in every function would be tedious (and might impact performance).

Instead, we can make a new type and put the validations in a function to create an instance of the type rather than repeating the validations everywhere. That way, it’s safe for functions to use the new type in their signatures and confidently use the values they receive. Listing 9-9 shows one way to define a Guess type that will only create an instance of Guess if the new function receives a value between 1 and 100:

# #![allow(unused_variables)]
#fn main() {
pub struct Guess {
    value: u32,
}

impl Guess {
    pub fn new(value: u32) -> Guess {
        if value < 1 || value > 100 {
            panic!("Guess value must be between 1 and 100, got {}.", value);
        }

        Guess {
            value
        }
    }

    pub fn value(&self) -> u32 {
        self.value
    }
}
#}

Listing 9-9: A Guess type that will only continue with values between 1 and 100

First, we define a struct named Guess that has a field named value that holds a u32. This is where the number will be stored.

Then we implement an associated function named new on Guess that creates instances of Guess values. The new function is defined to have one parameter named value of type u32 and to return a Guess. The code in the body of the new function tests value to make sure it’s between 1 and 100. If value doesn’t pass this test, we make a panic! call, which will alert the programmer who is writing the calling code that they have a bug they need to fix, because creating a Guess with a value outside this range would violate the contract that Guess::new is relying on. The conditions in which Guess::new might panic should be discussed in its public-facing API documentation; we’ll cover documentation conventions indicating the possibility of a panic! in the API documentation that you create in Chapter 14. If value does pass the test, we create a new Guess with its value field set to the value parameter and return the Guess.

Next, we implement a method named value that borrows self, doesn’t have any other parameters, and returns a u32. This kind of method is sometimes called a getter, because its purpose is to get some data from its fields and return it. This public method is necessary because the value field of the Guess struct is private. It’s important that the value field be private so code using the Guess struct is not allowed to set value directly: code outside the module must use the Guess::new function to create an instance of Guess, thereby ensuring there’s no way for a Guess to have a value that hasn’t been checked by the conditions in the Guess::new function.

A function that has a parameter or returns only numbers between 1 and 100 could then declare in its signature that it takes or returns a Guess rather than a u32 and wouldn’t need to do any additional checks in its body.

Rust’s error handling features are designed to help you write more robust code. The panic! macro signals that your program is in a state it can’t handle and lets you tell the process to stop instead of trying to proceed with invalid or incorrect values. The Result enum uses Rust’s type system to indicate that operations might fail in a way that your code could recover from. You can use Result to tell code that calls your code that it needs to handle potential success or failure as well. Using panic! and Result in the appropriate situations will make your code more reliable in the face of inevitable problems.

Now that you’ve seen useful ways that the standard library uses generics with the Option and Result enums, we’ll talk about how generics work and how you can use them in your code.

Every programming language has tools for effectively handling the duplication of concepts. In Rust, one such tool is generics. Generics are abstract stand-ins for concrete types or other properties. When we’re writing code, we can express the behavior of generics or how they relate to other generics without knowing what will be in their place when compiling and running the code.

Similar to the way a function takes parameters with unknown values to run the same code on multiple concrete values, functions can take parameters of some generic type instead of a concrete type, like i32 or String. In fact, we’ve already used generics in Chapter 6 with Option<T>, Chapter 8 with Vec<T> and HashMap<K, V>, and Chapter 9 with Result<T, E>. In this chapter, you’ll explore how to define your own types, functions, and methods with generics!

First, we’ll review how to extract a function to reduce code duplication. Next, we’ll use the same technique to make a generic function from two functions that only differ in the types of their parameters. We’ll also explain how to use generic types in struct and enum definitions.

Then you’ll learn how to use traits to define behavior in a generic way. You can then combine traits with generic types to constrain a generic type to only those types that have a particular behavior, as opposed to just any type.

Finally, we’ll discuss lifetimes, a variety of generics that give the compiler information about how references relate to each other. Lifetimes allow us to borrow values in many situations while still enabling the compiler to check that the references are valid.

Before diving into generics syntax, let’s first look at how to remove duplication that doesn’t involve generic types by extracting a function. Then we’ll apply this technique to extract a generic function! In the same way that you recognize duplicated code to extract into a function, you’ll start to recognize duplicated code that can use generics.

Consider a short program that finds the largest number in a list, as shown in Listing 10-1:

Filename: src/main.rs

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let mut largest = number_list[0];

    for number in number_list {
        if number > largest {
            largest = number;
        }
    }

    println!("The largest number is {}", largest);
#  assert_eq!(largest, 100);
}

Listing 10-1: Code to find the largest number in a list of numbers

This code stores a list of integers in the variable number_list and places the first number in the list in a variable named largest. Then it iterates through all the numbers in the list, and if the current number is greater than the number stored in largest, it replaces the number in that variable. However, if the current number is less than the largest number seen so far, the variable doesn’t change and the code moves on to the next number in the list. After considering all the numbers in the list, largest should hold the largest number, which in this case is 100.

To find the largest number in two different lists of numbers, we can duplicate the code in Listing 10-1 and use the same logic at two different places in the program, as shown in Listing 10-2:

Filename: src/main.rs

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let mut largest = number_list[0];

    for number in number_list {
        if number > largest {
            largest = number;
        }
    }

    println!("The largest number is {}", largest);

    let number_list = vec![102, 34, 6000, 89, 54, 2, 43, 8];

    let mut largest = number_list[0];

    for number in number_list {
        if number > largest {
            largest = number;
        }
    }

    println!("The largest number is {}", largest);
}

Listing 10-2: Code to find the largest number in two lists of numbers

Although this code works, duplicating code is tedious and error prone. We also have to update the code in multiple places to change it.

To eliminate this duplication, we can create an abstraction by defining a function that operates on any list of integers given to it in a parameter. This solution makes our code clearer and lets us express the concept of finding the largest number in a list abstractly.

In Listing 10-3, we extracted the code that finds the largest number into a function named largest. Unlike the code in Listing 10-1, which can find the largest number in only one particular list, this program can find the largest number in two different lists:

Filename: src/main.rs

fn largest(list: &[i32]) -> i32 {
    let mut largest = list[0];

    for &item in list.iter() {
        if item > largest {
            largest = item;
        }
    }

    largest
}

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let result = largest(&number_list);
    println!("The largest number is {}", result);
#    assert_eq!(result, 100);

    let number_list = vec![102, 34, 6000, 89, 54, 2, 43, 8];

    let result = largest(&number_list);
    println!("The largest number is {}", result);
#    assert_eq!(result, 6000);
}

Listing 10-3: Abstracted code to find the largest number in two lists

The largest function has a parameter called list, which represents any concrete slice of i32 values that we might pass into the function. As a result, when we call the function, the code runs on the specific values that we pass in.

In sum, here are the steps we took to change the code from Listing 10-2 to Listing 10-3:

1. Identify duplicate code.
2. Extract the duplicate code into the body of the function, and specify the inputs and return values of that code in the function signature.
3. Update the two instances of duplicated code to call the function instead.

Next, we’ll use these same steps with generics to reduce code duplication in different ways. In the same way that the function body can operate on an abstract list instead of specific values, generics allow code to operate on abstract types.

For example, say we had two functions: one that finds the largest item in a slice of i32 values and one that finds the largest item in a slice of char values. How would we eliminate that duplication? Let’s find out!

We can use generics to create definitions for items like function signatures or structs, which we can then use with many different concrete data types. Let’s first look at how to define functions, structs, enums, and methods using generics. Then we’ll discuss how generics affect code performance.

When defining a function that uses generics, we place the generics in the signature of the function where we would usually specify the data types of the parameters and return value. Doing so makes our code more flexible and provides more functionality to callers of our function while preventing code duplication.

Continuing with our largest function, Listing 10-4 shows two functions that both find the largest value in a slice:

Filename: src/main.rs

fn largest_i32(list: &[i32]) -> i32 {
    let mut largest = list[0];

    for &item in list.iter() {
        if item > largest {
            largest = item;
        }
    }

    largest
}

fn largest_char(list: &[char]) -> char {
    let mut largest = list[0];

    for &item in list.iter() {
        if item > largest {
            largest = item;
        }
    }

    largest
}

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let result = largest_i32(&number_list);
    println!("The largest number is {}", result);
#    assert_eq!(result, 100);

    let char_list = vec!['y', 'm', 'a', 'q'];

    let result = largest_char(&char_list);
    println!("The largest char is {}", result);
#    assert_eq!(result, 'y');
}

Listing 10-4: Two functions that differ only in their names and the types in their signatures

The largest_i32 function is the one we extracted in Listing 10-3 that finds the largest i32 in a slice. The largest_char function finds the largest char in a slice. The function bodies have the same code, so let’s eliminate the duplication by introducing a generic type parameter in a single function.

To parameterize the types in the new function we’ll define, we need to name the type parameter, just like we do for the value parameters to a function. You can use any identifier as a type parameter name. But we’ll use T because, by convention, parameter names in Rust are short, often just a letter, and Rust’s type naming convention is CamelCase. Short for “type,” T is the default choice of most Rust programmers.

When we use a parameter in the body of the function, we have to declare the parameter name in the signature so that the compiler knows what that name means. Similarly, when we use a type parameter name in a function signature, we have to declare the type parameter name before we use it. To define the generic largest function, place type name declarations inside angle brackets (<>) between the name of the function and the parameter list, like this:

fn largest<T>(list: &[T]) -> T {

We read this definition as: the function largest is generic over some type T. This function has one parameter named list, which is a slice of values of type T. The largest function will return a value of the same type T.

Listing 10-5 shows the combined largest function definition using the generic data type in its signature. The listing also shows how we can call the function with either a slice of i32 values or char values. Note that this code won’t compile yet, but we’ll fix it later in this chapter.

Filename: src/main.rs

fn largest<T>(list: &[T]) -> T {
    let mut largest = list[0];

    for &item in list.iter() {
        if item > largest {
            largest = item;
        }
    }

    largest
}

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let result = largest(&number_list);
    println!("The largest number is {}", result);

    let char_list = vec!['y', 'm', 'a', 'q'];

    let result = largest(&char_list);
    println!("The largest char is {}", result);
}

Listing 10-5: A definition of the largest function that uses generic type parameters but doesn’t compile yet

If we compile this code right now, we’ll get this error:

error[E0369]: binary operation `>` cannot be applied to type `T`
 --> src/main.rs:5:12
  |
5 |         if item > largest {
  |            ^^^^^^^^^^^^^^
  |
  = note: an implementation of `std::cmp::PartialOrd` might be missing for `T`

The note mentions std::cmp::PartialOrd, which is a trait. We’ll talk about traits in the next section. For now, this error states that the body of largest won’t work for all possible types that T could be. Because we want to compare values of type T in the body, we can only use types whose values can be ordered. To enable comparisons, the standard library has the std::cmp::PartialOrd trait that you can implement on types (see Appendix C, “Derivable Traits,” for more on this trait). You’ll learn how to specify that a generic type has a particular trait in the “Trait Bounds” section, but let’s first explore other ways of using generic type parameters.

We can also define structs to use a generic type parameter in one or more fields using the <> syntax. Listing 10-6 shows how to define a Point<T> struct to hold x and y coordinate values of any type:

Filename: src/main.rs

struct Point<T> {
    x: T,
    y: T,
}

fn main() {
    let integer = Point { x: 5, y: 10 };
    let float = Point { x: 1.0, y: 4.0 };
}

Listing 10-6: A Point<T> struct that holds x and y values of type T

The syntax for using generics in struct definitions is similar to that used in function definitions. First, we declare the name of the type parameter inside angle brackets just after the name of the struct. Then we can use the generic type in the struct definition where we would otherwise specify concrete data types.

Note that because we’ve only used one generic type to define Point<T>, this says that the Point<T> struct is generic over some type T, and the fields x and y are both that same type, whatever that type may be. This means that if we create an instance of a Point<T> that has values of different types, as in Listing 10-7, our code won’t compile:

Filename: src/main.rs

struct Point<T> {
    x: T,
    y: T,
}

fn main() {
    let wont_work = Point { x: 5, y: 4.0 };
}

Listing 10-7: The fields x and y must be the same type because both have the same generic data type T

In this example, when we assign the integer value 5 to x, we let the compiler know that the generic type T will be an integer for this instance of Point<T>. Then when we specify 4.0 for y, which we’ve defined to have the same type as x, we’ll get a type mismatch error like this:

error[E0308]: mismatched types
 --> src/main.rs:7:38
  |
7 |     let wont_work = Point { x: 5, y: 4.0 };
  |                                      ^^^ expected integral variable, found
floating-point variable
  |
  = note: expected type `{integer}`
             found type `{float}`

To define a Point struct where x and y are both generics but could have different types, we can use multiple generic type parameters. For example, in Listing 10-8, we can change the definition of Point to be generic over types T and U where x is of type T and y is of type U:

Filename: src/main.rs

struct Point<T, U> {
    x: T,
    y: U,
}

fn main() {
    let both_integer = Point { x: 5, y: 10 };
    let both_float = Point { x: 1.0, y: 4.0 };
    let integer_and_float = Point { x: 5, y: 4.0 };
}

Listing 10-8: A Point<T, U> generic over two types so that x and y can be values of different types

Now all the instances of Point shown are allowed! You can use as many generic type parameters in a definition as you want, but using more than a few makes your code hard to read. When you need lots of generic types in your code, it could indicate that your code needs restructuring into smaller pieces.

As we did with structs, we can define enums to hold generic data types in their variants. Let’s take another look at the Option<T> enum that the standard library provides that we used in Chapter 6:

# #![allow(unused_variables)]
#fn main() {
enum Option<T> {
    Some(T),
    None,
}
#}

This definition should now make more sense to you. As you can see, Option<T> is an enum that is generic over type T and has two variants: Some, which holds one value of type T, and a None variant that doesn’t hold any value. By using the Option<T> enum, we can express the abstract concept of having an optional value, and because Option<T> is generic, we can use this abstraction no matter what the type of the optional value is.

Enums can use multiple generic types as well. The definition of the Result enum that we used in Chapter 9 is one example:

# #![allow(unused_variables)]
#fn main() {
enum Result<T, E> {
    Ok(T),
    Err(E),
}
#}

The Result enum is generic over two types, T and E, and has two variants: Ok, which holds a value of type T, and Err, which holds a value of type E. This definition makes it convenient to use the Result enum anywhere we have an operation that might succeed (return a value of some type T) or fail (return an error of some type E). In fact, this is what we used to open a file in Listing 9-3 where T was filled in with the type std::fs::File when the file was opened successfully and E was filled in with the type std::io::Error when there were problems opening the file.

When you recognize situations in your code with multiple struct or enum definitions that differ only in the types of the values they hold, you can avoid duplication by using generic types instead.

As we did in Chapter 5, we can implement methods on structs and enums that have generic types in their definitions. Listing 10-9 shows the Point<T> struct we defined in Listing 10-6 with a method named x implemented on it:

Filename: src/main.rs

struct Point<T> {
    x: T,
    y: T,
}

impl<T> Point<T> {
    fn x(&self) -> &T {
        &self.x
    }
}

fn main() {
    let p = Point { x: 5, y: 10 };

    println!("p.x = {}", p.x());
}

Listing 10-9: Implementing a method named x on the Point<T> struct that will return a reference to the x field of type T

Here, we’ve defined a method named x on Point<T> that returns a reference to the data in the field x.

Note that we have to declare T just after impl so we can use it to specify that we’re implementing methods on the type Point<T>. By declaring T as a generic type after impl, Rust can identify that the type in the angle brackets in Point is a generic type rather than a concrete type.

We could, for example, implement methods only on Point<f32> instances rather than on Point<T> instances with any generic type. In Listing 10-10 we use the concrete type f32, meaning we don’t declare any types after impl:

# #![allow(unused_variables)]
#fn main() {
# struct Point<T> {
#     x: T,
#     y: T,
# }
#
impl Point<f32> {
    fn distance_from_origin(&self) -> f32 {
        (self.x.powi(2) + self.y.powi(2)).sqrt()
    }
}
#}

Listing 10-10: An impl block that only applies to a struct with a particular concrete type for the generic type parameter T

This code means the type Point<f32> will have a method named distance_from_origin, and other instances of Point<T> where T is not of type f32 will not have this method defined. The method measures how far our point is from the point at coordinates (0.0, 0.0) and uses mathematical operations that are only available for floating point types.

Generic type parameters in a struct definition aren’t always the same as those you use in that struct’s method signatures. For example, Listing 10-11 defines the method mixup on the Point<T, U> struct from Listing 10-8. The method takes another Point as a parameter, which might have different types than the self Point we’re calling mixup on. The method creates a new Point instance with the x value from the self Point (of type T) and the y value from the passed-in Point (of type W):

Filename: src/main.rs

struct Point<T, U> {
    x: T,
    y: U,
}

impl<T, U> Point<T, U> {
    fn mixup<V, W>(self, other: Point<V, W>) -> Point<T, W> {
        Point {
            x: self.x,
            y: other.y,
        }
    }
}

fn main() {
    let p1 = Point { x: 5, y: 10.4 };
    let p2 = Point { x: "Hello", y: 'c'};

    let p3 = p1.mixup(p2);

    println!("p3.x = {}, p3.y = {}", p3.x, p3.y);
}

Listing 10-11: Methods that use different generic types than their struct’s definition

In main, we’ve defined a Point that has an i32 for x (with value 5) and an f64 for y (with value 10.4). The p2 variable is a Point struct that has a string slice for x (with value "Hello") and a char for y (with value c). Calling mixup on p1 with the argument p2 gives us p3, which will have an i32 for x, because x came from p1. The p3 variable will have a char for y, because y came from p2. The println! macro call will print p3.x = 5, p3.y = c.

The purpose of this example is to demonstrate a situation in which some generic parameters are declared with impl and some are declared with the method definition. Here, the generic parameters T and U are declared after impl, because they go with the struct definition. The generic parameters V and W are declared after fn mixup, because they’re only relevant to the method.

You might be wondering whether there is a runtime cost when you’re using generic type parameters. The good news is that Rust implements generics in such a way that your code doesn’t run any slower using generic types than it would with concrete types.

Rust accomplishes this by performing monomorphization of the code that is using generics at compile time. Monomorphization is the process of turning generic code into specific code by filling in the concrete types that are used when compiled.

In this process, the compiler does the opposite of the steps we used to create the generic function in Listing 10-5: the compiler looks at all the places where generic code is called and generates code for the concrete types the generic code is called with.

Let’s look at how this works with an example that uses the standard library’s Option<T> enum:

# #![allow(unused_variables)]
#fn main() {
let integer = Some(5);
let float = Some(5.0);
#}

When Rust compiles this code, it performs monomorphization. During that process, the compiler reads the values that have been used in the instances of Option<T> and identifies two kinds of Option<T>: one is i32 and the other is f64. As such, it expands the generic definition of Option<T> into Option_i32 and Option_f64, thereby replacing the generic definition with the specific ones.

The monomorphized version of the code looks like the following. The generic Option<T> is replaced with the specific definitions created by the compiler:

Filename: src/main.rs

enum Option_i32 {
    Some(i32),
    None,
}

enum Option_f64 {
    Some(f64),
    None,
}

fn main() {
    let integer = Option_i32::Some(5);
    let float = Option_f64::Some(5.0);
}

Because Rust compiles generic code into code that specifies the type in each instance, we pay no runtime cost for using generics. When the code runs, it performs just like it would if we had duplicated each definition by hand. The process of monomorphization makes Rust’s generics extremely efficient at runtime.

A trait tells the Rust compiler about functionality a particular type has and can share with other types. We can use traits to define shared behavior in an abstract way. We can use trait bounds to specify that a generic can be any type that has certain behavior.

Note: Traits are similar to a feature often called interfaces in other languages, although with some differences.

A type’s behavior consists of the methods we can call on that type. Different types share the same behavior if we can call the same methods on all of those types. Trait definitions are a way to group method signatures together to define a set of behaviors necessary to accomplish some purpose.

For example, let’s say we have multiple structs that hold various kinds and amounts of text: a NewsArticle struct that holds a news story filed in a particular location and a Tweet that can have at most 280 characters along with metadata that indicates whether it was a new tweet, a retweet, or a reply to another tweet.

We want to make a media aggregator library that can display summaries of data that might be stored in a NewsArticle or Tweet instance. To do this, we need a summary from each type, and we need to request that summary by calling a summarize method on an instance. Listing 10-12 shows the definition of a Summary trait that expresses this behavior:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub trait Summary {
    fn summarize(&self) -> String;
}
#}

Listing 10-12: Definition of a Summary trait that consists of the behavior provided by a summarize method

Here, we declare a trait using the trait keyword and then the trait’s name, which is Summary in this case. Inside the curly brackets we declare the method signatures that describe the behaviors of the types that implement this trait, which in this case is fn summarize(&self) -> String.

After the method signature, instead of providing an implementation within curly brackets, we use a semicolon. Each type implementing this trait must provide its own custom behavior for the body of the method. The compiler will enforce that any type that has the Summary trait will have the method summarize defined with this signature exactly.

A trait can have multiple methods in its body: the method signatures are listed one per line and each line ends in a semicolon.

Now that we’ve defined the desired behavior using the Summary trait, we can implement it on the types in our media aggregator. Listing 10-13 shows an implementation of the Summary trait on the NewsArticle struct that uses the headline, the author, and the location to create the return value of summarize. For the Tweet struct, we define summarize as the username followed by the entire text of the tweet, assuming that tweet content is already limited to 280 characters.

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
# pub trait Summary {
#     fn summarize(&self) -> String;
# }
#
pub struct NewsArticle {
    pub headline: String,
    pub location: String,
    pub author: String,
    pub content: String,
}

impl Summary for NewsArticle {
    fn summarize(&self) -> String {
        format!("{}, by {} ({})", self.headline, self.author, self.location)
    }
}

pub struct Tweet {
    pub username: String,
    pub content: String,
    pub reply: bool,
    pub retweet: bool,
}

impl Summary for Tweet {
    fn summarize(&self) -> String {
        format!("{}: {}", self.username, self.content)
    }
}
#}

Listing 10-13: Implementing the Summary trait on the NewsArticle and Tweet types

Implementing a trait on a type is similar to implementing regular methods. The difference is that after impl, we put the trait name that we want to implement, then use the for keyword, and then specify the name of the type we want to implement the trait for. Within the impl block, we put the method signatures that the trait definition has defined. Instead of adding a semicolon after each signature, we use curly brackets and fill in the method body with the specific behavior that we want the methods of the trait to have for the particular type.

After implementing the trait, we can call the methods on instances of NewsArticle and Tweet in the same way we call regular methods, like this:

let tweet = Tweet {
    username: String::from("horse_ebooks"),
    content: String::from("of course, as you probably already know, people"),
    reply: false,
    retweet: false,
};

println!("1 new tweet: {}", tweet.summarize());

This code prints 1 new tweet: horse_ebooks: of course, as you probably already know, people.

Note that because we defined the Summary trait and the NewsArticle and Tweet types in the same lib.rs in Listing 10-13, they’re all in the same scope. Let’s say this lib.rs is for a crate we’ve called aggregator, and someone else wants to use our crate’s functionality to implement the Summary trait on a struct defined within their library’s scope. They would need to import the trait into their scope first. They would do so by specifying use aggregator::Summary;, which then enables them to implement Summary for their type. The Summary trait would also need to be a public trait for another crate to implement it, which it is because we put the pub keyword before trait in Listing 10-12.

One restriction to note with trait implementations is that we can implement a trait on a type only if either the trait or the type is local to your crate. For example, we can implement standard library traits like Display on a custom type like Tweet as part of our aggregator crate functionality, because the type Tweet is local to our aggregator crate. We can also implement Summary on Vec<T> in our aggregator crate, because the trait Summary is local to our aggregator crate.

But we can’t implement external traits on external types. For example, we can’t implement the Display trait on Vec<T> within our aggregator crate, because Display and Vec<T> are defined in the standard library and aren’t local to our aggregator crate. This restriction is part of a property of programs called coherence, and more specifically the orphan rule, so named because the parent type is not present. This rule ensures that other people’s code can’t break your code and vice versa. Without the rule, two crates could implement the same trait for the same type, and Rust wouldn’t know which implementation to use.

Sometimes it’s useful to have default behavior for some or all of the methods in a trait instead of requiring implementations for all methods on every type. Then, as we implement the trait on a particular type, we can keep or override each method’s default behavior.

Listing 10-14 shows how to specify a default string for the summarize method of the Summary trait instead of only defining the method signature, like we did in Listing 10-12:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub trait Summary {
    fn summarize(&self) -> String {
        String::from("(Read more...)")
    }
}
#}

Listing 10-14: Definition of a Summary trait with a default implementation of the summarize method

To use a default implementation to summarize instances of NewsArticle instead of defining a custom implementation, we specify an empty impl block with impl Summary for NewsArticle {}.

Even though we’re no longer defining the summarize method on NewsArticle directly, we’ve provided a default implementation and specified that NewsArticle implements the Summary trait. As a result, we can still call the summarize method on an instance of NewsArticle, like this:

let article = NewsArticle {
    headline: String::from("Penguins win the Stanley Cup Championship!"),
    location: String::from("Pittsburgh, PA, USA"),
    author: String::from("Iceburgh"),
    content: String::from("The Pittsburgh Penguins once again are the best
    hockey team in the NHL."),
};

println!("New article available! {}", article.summarize());

This code prints New article available! (Read more...).

Creating a default implementation for summarize doesn’t require us to change anything about the implementation of Summary on Tweet in Listing 10-13. The reason is that the syntax for overriding a default implementation is the same as the syntax for implementing a trait method that doesn’t have a default implementation.

Default implementations can call other methods in the same trait, even if those other methods don’t have a default implementation. In this way, a trait can provide a lot of useful functionality and only require implementors to specify a small part of it. For example, we could define the Summary trait to have a summarize_author method whose implementation is required, and then define a summarize method that has a default implementation that calls the summarize_author method:

# #![allow(unused_variables)]
#fn main() {
pub trait Summary {
    fn summarize_author(&self) -> String;

    fn summarize(&self) -> String {
        format!("(Read more from {}...)", self.summarize_author())
    }
}
#}

To use this version of Summary, we only need to define summarize_author when we implement the trait on a type:

impl Summary for Tweet {
    fn summarize_author(&self) -> String {
        format!("@{}", self.username)
    }
}

After we define summarize_author, we can call summarize on instances of the Tweet struct, and the default implementation of summarize will call the definition of summarize_author that we’ve provided. Because we’ve implemented summarize_author, the Summary trait has given us the behavior of the summarize method without requiring us to write any more code.

let tweet = Tweet {
    username: String::from("horse_ebooks"),
    content: String::from("of course, as you probably already know, people"),
    reply: false,
    retweet: false,
};

println!("1 new tweet: {}", tweet.summarize());

This code prints 1 new tweet: (Read more from @horse_ebooks...).

Note that it isn’t possible to call the default implementation from an overriding implementation of that same method.

Now that you know how to define traits and implement those traits on types, we can explore how to use traits with generic type parameters. We can use trait bounds to constrain generic types to ensure the type will be limited to those that implement a particular trait and behavior.

For example, in Listing 10-13, we implemented the Summary trait on the types NewsArticle and Tweet. We can define a function notify that calls the summarize method on its parameter item, which is of the generic type T. To be able to call summarize on item without getting an error telling us that the generic type T doesn’t implement the method summarize, we can use trait bounds on T to specify that item must be of a type that implements the Summary trait:

pub fn notify<T: Summary>(item: T) {
    println!("Breaking news! {}", item.summarize());
}

We place trait bounds with the declaration of the generic type parameter, after a colon and inside angle brackets. Because of the trait bound on T, we can call notify and pass in any instance of NewsArticle or Tweet. Code that calls the function with any other type, like a String or an i32, won’t compile, because those types don’t implement Summary.

We can specify multiple trait bounds on a generic type using the + syntax. For example, to use display formatting on the type T in a function as well as the summarize method, we can use T: Summary + Display to say T can be any type that implements Summary and Display.

However, there are downsides to using too many trait bounds. Each generic has its own trait bounds; so functions with multiple generic type parameters can have lots of trait bound information between a function’s name and its parameter list, making the function signature hard to read. For this reason, Rust has alternate syntax for specifying trait bounds inside a where clause after the function signature. So instead of writing this:

fn some_function<T: Display + Clone, U: Clone + Debug>(t: T, u: U) -> i32 {

we can use a where clause, like this:

fn some_function<T, U>(t: T, u: U) -> i32
    where T: Display + Clone,
          U: Clone + Debug
{

This function’s signature is less cluttered in that the function name, parameter list, and return type are close together, similar to a function without lots of trait bounds.

Now that you know how to specify the behavior you want to use using the generic type parameter’s bounds, let’s return to Listing 10-5 to fix the definition of the largest function that uses a generic type parameter! Last time we tried to run that code, we received this error:

error[E0369]: binary operation `>` cannot be applied to type `T`
 --> src/main.rs:5:12
  |
5 |         if item > largest {
  |            ^^^^^^^^^^^^^^
  |
  = note: an implementation of `std::cmp::PartialOrd` might be missing for `T`

In the body of largest we wanted to compare two values of type T using the greater-than (>) operator. Because that operator is defined as a default method on the standard library trait std::cmp::PartialOrd, we need to specify PartialOrd in the trait bounds for T so the largest function can work on slices of any type that we can compare. We don’t need to bring PartialOrd into scope because it’s in the prelude. Change the signature of largest to look like this:

fn largest<T: PartialOrd>(list: &[T]) -> T {

This time when we compile the code, we get a different set of errors:

error[E0508]: cannot move out of type `[T]`, a non-copy slice
 --> src/main.rs:2:23
  |
2 |     let mut largest = list[0];
  |                       ^^^^^^^
  |                       |
  |                       cannot move out of here
  |                       help: consider using a reference instead: `&list[0]`

error[E0507]: cannot move out of borrowed content
 --> src/main.rs:4:9
  |
4 |     for &item in list.iter() {
  |         ^----
  |         ||
  |         |hint: to prevent move, use `ref item` or `ref mut item`
  |         cannot move out of borrowed content

The key line in this error is cannot move out of type [T], a non-copy slice. With our non-generic versions of the largest function, we were only trying to find the largest i32 or char. As discussed in “Stack-Only Data: Copy” section in Chapter 4, types like i32 and char that have a known size can be stored on the stack, so they implement the Copy trait. But when we made the largest function generic, the list parameter could have types in it that don’t implement the Copy trait. Consequently, we wouldn’t be able to move the value out of list[0] and into the largest variable, resulting in this error.

To call this code with only those types that implement the Copy trait, we can add Copy to the trait bounds of T! Listing 10-15 shows the complete code of a generic largest function that will compile as long as the types of the values in the slice that we pass into the function implement the PartialOrd and Copy traits, like i32 and char do:

Filename: src/main.rs

fn largest<T: PartialOrd + Copy>(list: &[T]) -> T {
    let mut largest = list[0];

    for &item in list.iter() {
        if item > largest {
            largest = item;
        }
    }

    largest
}

fn main() {
    let number_list = vec![34, 50, 25, 100, 65];

    let result = largest(&number_list);
    println!("The largest number is {}", result);

    let char_list = vec!['y', 'm', 'a', 'q'];

    let result = largest(&char_list);
    println!("The largest char is {}", result);
}

Listing 10-15: A working definition of the largest function that works on any generic type that implements the PartialOrd and Copy traits

If we don’t want to restrict the largest function to the types that implement the Copy trait, we could specify that T has the trait bound Clone instead of Copy. Then we could clone each value in the slice when we want the largest function to have ownership. Using the clone function means we’re potentially making more heap allocations in the case of types that own heap data like String, and heap allocations can be slow if we’re working with large amounts of data.

Another way we could implement largest is for the function to return a reference to a T value in the slice. If we change the return type to &T instead of T, thereby changing the body of the function to return a reference, we wouldn’t need the Clone or Copy trait bounds and we could avoid heap allocations. Try implementing these alternate solutions on your own!

By using a trait bound with an impl block that uses generic type parameters, we can implement methods conditionally for types that implement the specified traits. For example, the type Pair<T> in Listing 10-16 always implements the new function. But Pair<T> only implements the cmp_display method if its inner type T implements the PartialOrd trait that enables comparison and the Display trait that enables printing:

# #![allow(unused_variables)]
#fn main() {
use std::fmt::Display;

struct Pair<T> {
    x: T,
    y: T,
}

impl<T> Pair<T> {
    fn new(x: T, y: T) -> Self {
        Self {
            x,
            y,
        }
    }
}

impl<T: Display + PartialOrd> Pair<T> {
    fn cmp_display(&self) {
        if self.x >= self.y {
            println!("The largest member is x = {}", self.x);
        } else {
            println!("The largest member is y = {}", self.y);
        }
    }
}
#}

Listing 10-16: Conditionally implement methods on a generic type depending on trait bounds

We can also conditionally implement a trait for any type that implements another trait. Implementations of a trait on any type that satisfies the trait bounds are called blanket implementations and are extensively used in the Rust standard library. For example, the standard library implements the ToString trait on any type that implements the Display trait. The impl block in the standard library looks similar to this code:

impl<T: Display> ToString for T {
    // --snip--
}

Because the standard library has this blanket implementation, we can call the to_string method defined by the ToString trait on any type that implements the Display trait. For example, we can turn integers into their corresponding String values like this because integers implement Display:

# #![allow(unused_variables)]
#fn main() {
let s = 3.to_string();
#}

Blanket implementations appear in the documentation for the trait in the “Implementors” section.

Traits and trait bounds let us write code that uses generic type parameters to reduce duplication but also specify to the compiler that we want the generic type to have particular behavior. The compiler can then use the trait bound information to check that all the concrete types used with our code provide the correct behavior. In dynamically typed languages, we would get an error at runtime if we called a method on a type that the type didn’t implement. But Rust moves these errors to compile time so we’re forced to fix the problems before our code is even able to run. Additionally, we don’t have to write code that checks for behavior at runtime because we’ve already checked at compile time. Doing so improves performance without having to give up the flexibility of generics.

Another kind of generic that we’ve already been using is called lifetimes. Rather than ensuring that a type has the behavior we want, lifetimes ensure that references are valid as long as we need them to be. Let’s look at how lifetimes do that.

One detail we didn’t discuss in the “References and Borrowing” section in Chapter 4 is that every reference in Rust has a lifetime, which is the scope for which that reference is valid. Most of the time lifetimes are implicit and inferred, just like most of the time types are inferred. We must annotate types when multiple types are possible. In a similar way, we must annotate lifetimes when the lifetimes of references could be related in a few different ways. Rust requires us to annotate the relationships using generic lifetime parameters to ensure the actual references used at runtime will definitely be valid.

The concept of lifetimes is somewhat different from tools in other programming languages, arguably making lifetimes Rust’s most distinctive feature. Although we won’t cover lifetimes in their entirety in this chapter, we’ll discuss common ways you might encounter lifetime syntax so you can become familiar with the concepts. See the “Advanced Lifetimes” section in Chapter 19 for more detailed information.

The main aim of lifetimes is to prevent dangling references, which cause a program to reference data other than the data it’s intended to reference. Consider the program in Listing 10-17, which has an outer scope and an inner scope:

{
    let r;

    {
        let x = 5;
        r = &x;
    }

    println!("r: {}", r);
}

Listing 10-17: An attempt to use a reference whose value has gone out of scope

Note: The example in Listing 10-17 and the next few examples declare variables without giving them an initial value, so the variable name exists in the outer scope. At first glance, this might appear to be in conflict with Rust having no null values. However, if we try to use a variable before giving it a value, we’ll get a compile time error, which shows that Rust indeed does not allow null values.

The outer scope declares a variable named r with no initial value, and the inner scope declares a variable named x with the initial value of 5. Inside the inner scope, we attempt to set the value of r as a reference to x. Then the inner scope ends, and we attempt to print the value in r. This code won’t compile because the value r is referring to has gone out of scope before we try to use it. Here is the error message:

error[E0597]: `x` does not live long enough
  --> src/main.rs:7:5
   |
6  |         r = &x;
   |              - borrow occurs here
7  |     }
   |     ^ `x` dropped here while still borrowed
...
10 | }
   | - borrowed value needs to live until here

The variable x doesn’t “live long enough.” The reason is that x will be out of scope when the inner scope ends on line 7. But r is still valid for the outer scope; because its scope is larger, we say that it “lives longer.” If Rust allowed this code to work, r would be referencing memory that was deallocated when x went out of scope, and anything we tried to do with r wouldn’t work correctly. So how does Rust determine that this code is invalid? It uses a borrow checker.

The Rust compiler has a borrow checker that compares scopes to determine that all borrows are valid. Listing 10-18 shows the same code as Listing 10-17 but with annotations showing the lifetimes of the variables:

{
    let r;                // ---------+-- 'a
                          //          |
    {                     //          |
        let x = 5;        // -+-- 'b  |
        r = &x;           //  |       |
    }                     // -+       |
                          //          |
    println!("r: {}", r); //          |
}                         // ---------+

Listing 10-18: Annotations of the lifetimes of r and x, named 'a and 'b, respectively

Here, we’ve annotated the lifetime of r with 'a and the lifetime of x with 'b. As you can see, the inner 'b block is much smaller than the outer 'a lifetime block. At compile time, Rust compares the size of the two lifetimes and sees that r has a lifetime of 'a but that it refers to memory with a lifetime of 'b. The program is rejected because 'b is shorter than 'a: the subject of the reference doesn’t live as long as the reference.

Listing 10-19 fixes the code so it doesn’t have a dangling reference and compiles without any errors:

# #![allow(unused_variables)]
#fn main() {
{
    let x = 5;            // ----------+-- 'b
                          //           |
    let r = &x;           // --+-- 'a  |
                          //   |       |
    println!("r: {}", r); //   |       |
                          // --+       |
}                         // ----------+
#}

Listing 10-19: A valid reference because the data has a longer lifetime than the reference

Here, x has the lifetime 'b, which in this case is larger than 'a. This means r can reference x because Rust knows that the reference in r will always be valid while x is valid.

Now that you know where the lifetimes of references are and how Rust analyzes lifetimes to ensure references will always be valid, let’s explore generic lifetimes of parameters and return values in the context of functions.

Let’s write a function that returns the longer of two string slices. This function will take two string slices and return a string slice. After we’ve implemented the longest function, the code in Listing 10-20 should print The longest string is abcd:

Filename: src/main.rs

fn main() {
    let string1 = String::from("abcd");
    let string2 = "xyz";

    let result = longest(string1.as_str(), string2);
    println!("The longest string is {}", result);
}

Listing 10-20: A main function that calls the longest function to find the longest of two string slices

Note that we want the function to take string slices, which are references, because we don’t want the longest function to take ownership of its parameters. We want to allow the function to accept slices of a String (the type stored in the variable string1) as well as string literals (which is what variable string2 contains).

Refer to the “String Slices as Parameters” section in Chapter 4 for more discussion about why the parameters we use in Listing 10-20 are the ones we want.

If we try to implement the longest function as shown in Listing 10-21, it won’t compile:

Filename: src/main.rs

fn longest(x: &str, y: &str) -> &str {
    if x.len() > y.len() {
        x
    } else {
        y
    }
}

Listing 10-21: An implementation of the longest function that returns the longer of two string slices but does not yet compile

Instead, we get the following error that talks about lifetimes:

error[E0106]: missing lifetime specifier
 --> src/main.rs:1:33
  |
1 | fn longest(x: &str, y: &str) -> &str {
  |                                 ^ expected lifetime parameter
  |
  = help: this function's return type contains a borrowed value, but the
signature does not say whether it is borrowed from `x` or `y`

The help text reveals that the return type needs a generic lifetime parameter on it because Rust can’t tell whether the reference being returned refers to x or y. Actually, we don’t know either, because the if block in the body of this function returns a reference to x and the else block returns a reference to y!

When we’re defining this function, we don’t know the concrete values that will be passed into this function, so we don’t know whether the if case or the else case will execute. We also don’t know the concrete lifetimes of the references that will be passed in, so we can’t look at the scopes like we did in Listings 10-18 and 10-19 to determine that the reference we return will always be valid. The borrow checker can’t determine this either, because it doesn’t know how the lifetimes of x and y relate to the lifetime of the return value. To fix this error, we’ll add generic lifetime parameters that define the relationship between the references so the borrow checker can perform its analysis.

Lifetime annotations don’t change how long any of the references live. Just like functions can accept any type when the signature specifies a generic type parameter, functions can accept references with any lifetime by specifying a generic lifetime parameter. Lifetime annotations describe the relationships of the lifetimes of multiple references to each other without affecting the lifetimes.

Lifetime annotations have a slightly unusual syntax: the names of lifetime parameters must start with an apostrophe ' and are usually all lowercase and very short, like generic types. Most people use the name 'a. We place lifetime parameter annotations after the & of a reference, using a space to separate the annotation from the reference’s type.

Here are some examples: a reference to an i32 without a lifetime parameter, a reference to an i32 that has a lifetime parameter named 'a, and a mutable reference to an i32 that also has the lifetime 'a:

&i32        // a reference
&'a i32     // a reference with an explicit lifetime
&'a mut i32 // a mutable reference with an explicit lifetime

One lifetime annotation by itself doesn’t have much meaning because the annotations are meant to tell Rust how generic lifetime parameters of multiple references relate to each other. For example, let’s say we have a function with the parameter first that is a reference to an i32 with lifetime 'a. The function also has another parameter named second that is another reference to an i32 that also has the lifetime 'a. The lifetime annotations indicate that the references first and second must both live as long as that generic lifetime.

Now let’s examine lifetime annotations in the context of the longest function. As with generic type parameters, we need to declare generic lifetime parameters inside angle brackets between the function name and the parameter list. The constraint we want to express in this signature is that all the references in the parameters and the return value must have the same lifetime. We’ll name the lifetime 'a, and then add it to each reference, as shown in Listing 10-22:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
fn longest<'a>(x: &'a str, y: &'a str) -> &'a str {
    if x.len() > y.len() {
        x
    } else {
        y
    }
}
#}

Listing 10-22: The longest function definition specifying that all the references in the signature must have the same lifetime 'a

This code should compile and produce the result we want when we use it with the main function in Listing 10-20.

The function signature now tells Rust that for some lifetime 'a, the function takes two parameters, both of which are string slices that live at least as long as lifetime 'a. The function signature also tells Rust that the string slice returned from the function will live at least as long as lifetime 'a. These constraints are what we want Rust to enforce.

As discussed earlier, by specifying the lifetime parameters in this function signature, we’re not changing the lifetimes of any values passed in or returned. Instead, we’re specifying that the borrow checker should reject any values that don’t adhere to these constraints. Note that the longest function doesn’t need to know exactly how long x and y will live, only that some scope can be substituted for 'a that will satisfy this signature.

When annotating lifetimes in functions, the annotations go in the function signature, not in the function body. Rust can analyze the code within the function without any help. However, when a function has references to or from code outside that function, it becomes almost impossible for Rust to figure out the lifetimes of the parameters or return values on its own. The lifetimes might be different each time the function is called. This is why we need to annotate the lifetimes manually.

When we pass concrete references to longest, the concrete lifetime that is substituted for 'a is the part of the scope of x that overlaps with the scope of y. In other words, the generic lifetime 'a will get the concrete lifetime that is equal to the smaller of the lifetimes of x and y. Because we’ve annotated the returned reference with the same lifetime parameter 'a, the returned reference will also be valid for the length of the smaller of the lifetimes of x and y.

Let’s look at how the lifetime annotations restrict the longest function by passing in references that have different concrete lifetimes. Listing 10-23 is a straightforward example:

Filename: src/main.rs

# fn longest<'a>(x: &'a str, y: &'a str) -> &'a str {
#     if x.len() > y.len() {
#         x
#     } else {
#         y
#     }
# }
#
fn main() {
    let string1 = String::from("long string is long");

    {
        let string2 = String::from("xyz");
        let result = longest(string1.as_str(), string2.as_str());
        println!("The longest string is {}", result);
    }
}

Listing 10-23: Using the longest function with references to String values that have different concrete lifetimes

In this example, string1 is valid until the end of the outer scope, string2 is valid until the end of the inner scope, and result references something that is valid until the end of the inner scope. Run this code, and you’ll see that the borrow checker approves of this code; it will compile and print The longest string is long string is long.

Next, let’s try an example that shows that the lifetime of the reference in result must be the smaller lifetime of the two arguments. We’ll move the declaration of the result variable outside the inner scope but leave the assignment of the value to the result variable inside the scope with string2. Then we’ll move the println! that uses result outside the inner scope, after the inner scope has ended. The code in Listing 10-24 will not compile:

Filename: src/main.rs

fn main() {
    let string1 = String::from("long string is long");
    let result;
    {
        let string2 = String::from("xyz");
        result = longest(string1.as_str(), string2.as_str());
    }
    println!("The longest string is {}", result);
}

Listing 10-24: Attempting to use result after string2 has gone out of scope; the code won’t compile

When we try to compile this code, we’ll get this error:

error[E0597]: `string2` does not live long enough
  --> src/main.rs:15:5
   |
14 |         result = longest(string1.as_str(), string2.as_str());
   |                                            ------- borrow occurs here
15 |     }
   |     ^ `string2` dropped here while still borrowed
16 |     println!("The longest string is {}", result);
17 | }
   | - borrowed value needs to live until here

The error shows that for result to be valid for the println! statement, string2 would need to be valid until the end of the outer scope. Rust knows this because we annotated the lifetimes of the function parameters and return values using the same lifetime parameter 'a.

As humans, we can look at this code and see that string1 is longer than string2, and therefore result will contain a reference to string1. Because string1 has not gone out of scope yet, a reference to string1 will still be valid for the println! statement. However, the compiler can’t see that the reference is valid in this case. We’ve told Rust that the lifetime of the reference returned by the longest function is the same as the smaller of the lifetimes of the references passed in. Therefore, the borrow checker disallows the code in Listing 10-24 as possibly having an invalid reference.

Try designing more experiments that vary the values and lifetimes of the references passed in to the longest function and how the returned reference is used. Make hypotheses about whether or not your experiments will pass the borrow checker before you compile; then check to see if you’re right!

The way in which you need to specify lifetime parameters depends on what your function is doing. For example, if we changed the implementation of the longest function to always return the first parameter rather than the longest string slice, we wouldn’t need to specify a lifetime on the y parameter. The following code will compile:

Filename: src/main.rs

# #![allow(unused_variables)]
#fn main() {
fn longest<'a>(x: &'a str, y: &str) -> &'a str {
    x
}
#}

In this example, we’ve specified a lifetime parameter 'a for the parameter x and the return type, but not for the parameter y, because the lifetime of y does not have any relationship with the lifetime of x or the return value.

When returning a reference from a function, the lifetime parameter for the return type needs to match the lifetime parameter for one of the parameters. If the reference returned does not refer to one of the parameters, it must refer to a value created within this function, which would be a dangling reference because the value will go out of scope at the end of the function. Consider this attempted implementation of the longest function that won’t compile:

Filename: src/main.rs

fn longest<'a>(x: &str, y: &str) -> &'a str {
    let result = String::from("really long string");
    result.as_str()
}

Here, even though we’ve specified a lifetime parameter 'a for the return type, this implementation will fail to compile because the return value lifetime is not related to the lifetime of the parameters at all. Here is the error message we get:

error[E0597]: `result` does not live long enough
 --> src/main.rs:3:5
  |
3 |     result.as_str()
  |     ^^^^^^ does not live long enough
4 | }
  | - borrowed value only lives until here
  |
note: borrowed value must be valid for the lifetime 'a as defined on the
function body at 1:1...
 --> src/main.rs:1:1
  |
1 | / fn longest<'a>(x: &str, y: &str) -> &'a str {
2 | |     let result = String::from("really long string");
3 | |     result.as_str()
4 | | }
  | |_^

The problem is that result goes out of scope and gets cleaned up at the end of the longest function. We’re also trying to return a reference to result from the function. There is no way we can specify lifetime parameters that would change the dangling reference, and Rust won’t let us create a dangling reference. In this case, the best fix would be to return an owned data type rather than a reference so the calling function is then responsible for cleaning up the value.

Ultimately, lifetime syntax is about connecting the lifetimes of various parameters and return values of functions. Once they’re connected, Rust has enough information to allow memory-safe operations and disallow operations that would create dangling pointers or otherwise violate memory safety.

So far, we’ve only defined structs to hold owned types. It’s possible for structs to hold references, but in that case we would need to add a lifetime annotation on every reference in the struct’s definition. Listing 10-25 has a struct named ImportantExcerpt that holds a string slice:

Filename: src/main.rs

struct ImportantExcerpt<'a> {
    part: &'a str,
}

fn main() {
    let novel = String::from("Call me Ishmael. Some years ago...");
    let first_sentence = novel.split('.')
        .next()
        .expect("Could not find a '.'");
    let i = ImportantExcerpt { part: first_sentence };
}

Listing 10-25: A struct that holds a reference, so its definition needs a lifetime annotation

This struct has one field, part, that holds a string slice, which is a reference. As with generic data types, we declare the name of the generic lifetime parameter inside angle brackets after the name of the struct so we can use the lifetime parameter in the body of the struct definition. This annotation means an instance of ImportantExcerpt can’t outlive the reference it holds in its part field.

The main function here creates an instance of the ImportantExcerpt struct that holds a reference to the first sentence of the String owned by the variable novel. The data in novel exists before the ImportantExcerpt instance is created. In addition, novel doesn’t go out of scope until after the ImportantExcerpt goes out of scope, so the reference in the ImportantExcerpt instance is valid.

You’ve learned that every reference has a lifetime and that you need to specify lifetime parameters for functions or structs that use references. However, in Chapter 4 we had a function in the “String Slices” section, which is shown again in Listing 10-26, that compiled without lifetime annotations:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
fn first_word(s: &str) -> &str {
    let bytes = s.as_bytes();

    for (i, &item) in bytes.iter().enumerate() {
        if item == b' ' {
            return &s[0..i];
        }
    }

    &s[..]
}
#}

Listing 10-26: A function we defined in Chapter 4 that compiled without lifetime annotations, even though the parameter and return type are references

The reason this function compiles without lifetime annotations is historical: in early versions (pre-1.0) of Rust, this code wouldn’t have compiled because every reference needed an explicit lifetime. At that time, the function signature would have been written like this:

fn first_word<'a>(s: &'a str) -> &'a str {

After writing a lot of Rust code, the Rust team found that Rust programmers were entering the same lifetime annotations repeatedly in particular situations. These situations were predictable and followed a few deterministic patterns. The developers programmed these patterns into the compiler’s code so the borrow checker could infer the lifetimes in these situations and not need explicit annotations.

This piece of Rust history is relevant because it’s possible that more deterministic patterns will emerge and be added to the compiler. In the future, even fewer lifetime annotations might be required.

The patterns programmed into Rust’s analysis of references are called the lifetime elision rules. These aren’t rules for programmers to follow; they’re a set of particular cases that the compiler will consider, and if your code fits these cases, you don’t need to write the lifetimes explicitly.

The elision rules don’t provide full inference. If Rust deterministically applies the rules but there is still ambiguity as to what lifetimes the references have, the compiler won’t guess what the lifetime of the remaining references should be. In this case, instead of guessing, the compiler will give you an error that you can resolve by adding the lifetime annotations that specify how the references relate to each other.

Lifetimes on function or method parameters are called input lifetimes, and lifetimes on return values are called output lifetimes.

The compiler uses three rules to figure out what lifetimes references have when there aren’t explicit annotations. The first rule applies to input lifetimes, and the second and third rules apply to output lifetimes. If the compiler gets to the end of the three rules and there are still references for which it can’t figure out lifetimes, the compiler will stop with an error.

The first rule is that each parameter that is a reference gets its own lifetime parameter. In other words, a function with one parameter gets one lifetime parameter: fn foo<'a>(x: &'a i32); a function with two parameters gets two separate lifetime parameters: fn foo<'a, 'b>(x: &'a i32, y: &'b i32); and so on.

The second rule is if there is exactly one input lifetime parameter, that lifetime is assigned to all output lifetime parameters: fn foo<'a>(x: &'a i32) -> &'a i32.

The third rule is if there are multiple input lifetime parameters, but one of them is &self or &mut self because this is a method, the lifetime of self is assigned to all output lifetime parameters. This third rule makes methods much nicer to read and write because fewer symbols are necessary.

Let’s pretend we’re the compiler. We’ll apply these rules to figure out what the lifetimes of the references in the signature of the first_word function in Listing 10-26 are. The signature starts without any lifetimes associated with the references:

fn first_word(s: &str) -> &str {

Then the compiler applies the first rule, which specifies that each parameter gets its own lifetime. We’ll call it 'a as usual, so now the signature is:

fn first_word<'a>(s: &'a str) -> &str {

The second rule applies because there is exactly one input lifetime. The second rule specifies that the lifetime of the one input parameter gets assigned to the output lifetime, so the signature is now this:

fn first_word<'a>(s: &'a str) -> &'a str {

Now all the references in this function signature have lifetimes, and the compiler can continue its analysis without needing the programmer to annotate the lifetimes in this function signature.

Let’s look at another example, this time using the longest function that had no lifetime parameters when we started working with it in Listing 10-21:

fn longest(x: &str, y: &str) -> &str {

Let’s apply the first rule: each parameter gets its own lifetime. This time we have two parameters instead of one, so we have two lifetimes:

fn longest<'a, 'b>(x: &'a str, y: &'b str) -> &str {

You can see that the second rule doesn’t apply because there is more than one input lifetime. The third rule doesn’t apply either, because longest is a function rather than a method, so none of the parameters are self. After working through all three rules, we still haven’t figured out what the return type’s lifetime is. This is why we got an error trying to compile the code in Listing 10-21: the compiler worked through the lifetime elision rules but still couldn’t figure out all the lifetimes of the references in the signature.

Because the third rule really only applies in method signatures, we’ll look at lifetimes in that context next to see why the third rule means we don’t have to annotate lifetimes in method signatures very often.

When we implement methods on a struct with lifetimes, we use the same syntax as that of generic type parameters shown in Listing 10-11. Where we declare and use the lifetime parameters depends on whether they’re related to the struct fields or the method parameters and return values.

Lifetime names for struct fields always need to be declared after the impl keyword and then used after the struct’s name, because those lifetimes are part of the struct’s type.

In method signatures inside the impl block, references might be tied to the lifetime of references in the struct’s fields, or they might be independent. In addition, the lifetime elision rules often make it so that lifetime annotations aren’t necessary in method signatures. Let’s look at some examples using the struct named ImportantExcerpt that we defined in Listing 10-25.

First, we’ll use a method named level whose only parameter is a reference to self and whose return value is an i32, which is not a reference to anything:

# #![allow(unused_variables)]
#fn main() {
# struct ImportantExcerpt<'a> {
#     part: &'a str,
# }
#
impl<'a> ImportantExcerpt<'a> {
    fn level(&self) -> i32 {
        3
    }
}
#}

The lifetime parameter declaration after impl and use after the type name is required, but we’re not required to annotate the lifetime of the reference to self because of the first elision rule.

Here is an example where the third lifetime elision rule applies:

# #![allow(unused_variables)]
#fn main() {
# struct ImportantExcerpt<'a> {
#     part: &'a str,
# }
#
impl<'a> ImportantExcerpt<'a> {
    fn announce_and_return_part(&self, announcement: &str) -> &str {
        println!("Attention please: {}", announcement);
        self.part
    }
}
#}

There are two input lifetimes, so Rust applies the first lifetime elision rule and gives both &self and announcement their own lifetimes. Then, because one of the parameters is &self, the return type gets the lifetime of &self, and all lifetimes have been accounted for.

One special lifetime we need to discuss is 'static, which denotes the entire duration of the program. All string literals have the 'static lifetime, which we can annotate as follows: let s: &'static str = "I have a static lifetime.";

The text of this string is stored directly in the binary of your program, which is always available. Therefore, the lifetime of all string literals is 'static.

You might see suggestions to use the 'static lifetime in error messages. But before specifying 'static as the lifetime for a reference, think about whether the reference you have actually lives the entire lifetime of your program or not. You might consider whether you want it to live that long, even if it could. Most of the time, the problem results from attempting to create a dangling reference or a mismatch of the available lifetimes. In such cases, the solution is fixing those problems, not specifying the 'static lifetime.

Let’s briefly look at the syntax of specifying generic type parameters, trait bounds, and lifetimes all in one function!

# #![allow(unused_variables)]
#fn main() {
use std::fmt::Display;

fn longest_with_an_announcement<'a, T>(x: &'a str, y: &'a str, ann: T) -> &'a str
    where T: Display
{
    println!("Announcement! {}", ann);
    if x.len() > y.len() {
        x
    } else {
        y
    }
}
#}

This is the longest function from Listing 10-22 that returns the longer of two string slices. But now it has an extra parameter named ann of the generic type T, which can be filled in by any type that implements the Display trait as specified by the where clause. This extra parameter will be printed before the function compares the lengths of the string slices, which is why the Display trait bound is necessary. Because lifetimes are a type of generic, the declarations of the lifetime parameter 'a and the generic type parameter T go in the same list inside the angle brackets after the function name.

We covered a lot in this chapter! Now that you know about generic type parameters, traits and trait bounds, and generic lifetime parameters, you’re ready to write code without repetition yet works in many different situations. Generic type parameters let you apply the code to different types. Traits and trait bounds ensure that even though the types are generic, they’ll have the behavior the code needs. You learned how to use lifetime annotations to ensure that this flexible code won’t have any dangling references. And all of this analysis happens at compile time, which doesn’t affect runtime performance!

Believe it or not, there is much more to learn on the topics we discussed in this chapter: Chapter 17 discusses trait objects, which are another way to use traits. Chapter 19 covers more complex scenarios involving lifetime annotations as well as some advanced type system features. But in the next chapter, you’ll learn how to write tests in Rust so you can make sure your code is working the way it should.

In his 1972 essay “The Humble Programmer,” Edsger W. Dijkstra said that “Program testing can be a very effective way to show the presence of bugs, but it is hopelessly inadequate for showing their absence.” That doesn’t mean we shouldn’t try to test as much as we can!

Correctness in our programs is the extent to which our code does what we intend it to do. Rust is designed with a high degree of concern about the correctness of programs, but correctness is complex and not easy to prove. Rust’s type system shoulders a huge part of this burden, but the type system cannot catch every kind of incorrectness. As such, Rust includes support for writing automated software tests within the language.

As an example, say we write a function called add_two that adds 2 to whatever number is passed to it. This function’s signature accepts an integer as a parameter and returns an integer as a result. When we implement and compile that function, Rust does all the type checking and borrow checking that you’ve learned so far to ensure that, for instance, we aren’t passing a String value or an invalid reference to this function. But Rust can’t check that this function will do precisely what we intend, which is return the parameter plus 2 rather than, say, the parameter plus 10 or the parameter minus 50! That’s where tests come in.

We can write tests that assert, for example, that when we pass 3 to the add_two function, the returned value is 5. We can run these tests whenever we make changes to our code to make sure any existing correct behavior has not changed.

Testing is a complex skill: although we can’t cover every detail about how to write good tests in one chapter, we’ll discuss the mechanics of Rust’s testing facilities. We’ll talk about the annotations and macros available to you when writing your tests, the default behavior and options provided for running your tests, and how to organize tests into unit tests and integration tests.

Tests are Rust functions that verify that the non-test code is functioning in the expected manner. The bodies of test functions typically perform these three actions:

1. Set up any needed data or state.
2. Run the code you want to test.
3. Assert the results are what you expect.

Let’s look at the features Rust provides specifically for writing tests that take these actions, which include the test attribute, a few macros, and the should_panic attribute.

At its simplest, a test in Rust is a function that’s annotated with the test attribute. Attributes are metadata about pieces of Rust code; one example is the derive attribute we used with structs in Chapter 5. To change a function into a test function, add #[test] on the line before fn. When you run your tests with the cargo test command, Rust builds a test runner binary that runs the functions annotated with the test attribute and reports on whether each test function passes or fails.

In Chapter 7, we saw that when we make a new library project with Cargo, a test module with a test function in it is automatically generated for us. This module helps you start writing your tests so you don’t have to look up the exact structure and syntax of test functions every time you start a new project. You can add as many additional test functions and as many test modules as you want!

We’ll explore some aspects of how tests work by experimenting with the template test generated for us without actually testing any code. Then we’ll write some real-world tests that call some code that we’ve written and assert that its behavior is correct.

Let’s create a new library project called adder:

$ cargo new adder --lib
     Created library `adder` project
$ cd adder

The contents of the src/lib.rs file in your adder library should look like Listing 11-1:

Filename: src/lib.rs

# fn main() {}
#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
    }
}

Listing 11-1: The test module and function generated automatically by cargo new

For now, let’s ignore the top two lines and focus on the function to see how it works. Note the #[test] annotation before the fn line: this attribute indicates this is a test function, so the test runner knows to treat this function as a test. We could also have non-test functions in the tests module to help set up common scenarios or perform common operations, so we need to indicate which functions are tests by using the #[test] attribute.

The function body uses the assert_eq! macro to assert that 2 + 2 equals 4. This assertion serves as an example of the format for a typical test. Let’s run it to see that this test passes.

The cargo test command runs all tests in our project, as shown in Listing 11-2:

$ cargo test
   Compiling adder v0.1.0 (file:///projects/adder)
    Finished dev [unoptimized + debuginfo] target(s) in 0.22 secs
     Running target/debug/deps/adder-ce99bcc2479f4607

running 1 test
test tests::it_works ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

   Doc-tests adder

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Listing 11-2: The output from running the automatically generated test

Cargo compiled and ran the test. After the Compiling, Finished, and Running lines is the line running 1 test. The next line shows the name of the generated test function, called it_works, and the result of running that test, ok. The overall summary of running the tests appears next. The text test result: ok. means that all the tests passed, and the portion that reads 1 passed; 0 failed totals the number of tests that passed or failed.

Because we don’t have any tests we’ve marked as ignored, the summary shows 0 ignored. We also haven’t filtered the tests being run, so the end of the summary shows 0 filtered out. We’ll talk about ignoring and filtering out tests in the next section, “Controlling How Tests Are Run.”

The 0 measured statistic is for benchmark tests that measure performance. Benchmark tests are, as of this writing, only available in nightly Rust. See the documentation about benchmark tests to learn more.

The next part of the test output, which starts with Doc-tests adder, is for the results of any documentation tests. We don’t have any documentation tests yet, but Rust can compile any code examples that appear in our API documentation. This feature helps us keep our docs and our code in sync! We’ll discuss how to write documentation tests in the “Documentation Comments” section of Chapter 14. For now, we’ll ignore the Doc-tests output.

Let’s change the name of our test to see how that changes the test output. Change the it_works function to a different name, such as exploration, like so:

Filename: src/lib.rs

# fn main() {}
#[cfg(test)]
mod tests {
    #[test]
    fn exploration() {
        assert_eq!(2 + 2, 4);
    }
}

Then run cargo test again. The output now shows exploration instead of it_works:

running 1 test
test tests::exploration ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Let’s add another test, but this time we’ll make a test that fails! Tests fail when something in the test function panics. Each test is run in a new thread, and when the main thread sees that a test thread has died, the test is marked as failed. We talked about the simplest way to cause a panic in Chapter 9, which is to call the panic! macro. Enter the new test, another, so your src/lib.rs file looks like Listing 11-3:

Filename: src/lib.rs

# fn main() {}
#[cfg(test)]
mod tests {
    #[test]
    fn exploration() {
        assert_eq!(2 + 2, 4);
    }

    #[test]
    fn another() {
        panic!("Make this test fail");
    }
}

Listing 11-3: Adding a second test that will fail because we call the panic! macro

Run the tests again using cargo test. The output should look like Listing 11-4, which shows that our exploration test passed and another failed:

running 2 tests
test tests::exploration ... ok
test tests::another ... FAILED

failures:

---- tests::another stdout ----
    thread 'tests::another' panicked at 'Make this test fail', src/lib.rs:10:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    tests::another

test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

error: test failed

Listing 11-4: Test results when one test passes and one test fails

Instead of ok, the line test tests::another shows FAILED. Two new sections appear between the individual results and the summary: the first section displays the detailed reason for each test failure. In this case, another failed because it panicked at 'Make this test fail', which happened on line 10 in the src/lib.rs file. The next section lists just the names of all the failing tests, which is useful when there are lots of tests and lots of detailed failing test output. We can use the name of a failing test to run just that test to more easily debug it; we’ll talk more about ways to run tests in the “Controlling How Tests Are Run” section.

The summary line displays at the end: overall, our test result is FAILED. We had one test pass and one test fail.

Now that you’ve seen what the test results look like in different scenarios, let’s look at some macros other than panic! that are useful in tests.

The assert! macro, provided by the standard library, is useful when you want to ensure that some condition in a test evaluates to true. We give the assert! macro an argument that evaluates to a Boolean. If the value is true, assert! does nothing and the test passes. If the value is false, the assert! macro calls the panic! macro, which causes the test to fail. Using the assert! macro helps us check that our code is functioning in the way we intend.

In Chapter 5, Listing 5-15, we used a Rectangle struct and a can_hold method, which are repeated here in Listing 11-5. Let’s put this code in the src/lib.rs file and write some tests for it using the assert! macro.

Filename: src/lib.rs

# fn main() {}
#[derive(Debug)]
pub struct Rectangle {
    length: u32,
    width: u32,
}

impl Rectangle {
    pub fn can_hold(&self, other: &Rectangle) -> bool {
        self.length > other.length && self.width > other.width
    }
}

Listing 11-5: Using the Rectangle struct and its can_hold method from Chapter 5

The can_hold method returns a Boolean, which means it’s a perfect use case for the assert! macro. In Listing 11-6, we write a test that exercises the can_hold method by creating a Rectangle instance that has a length of 8 and a width of 7 and asserting that it can hold another Rectangle instance that has a length of 5 and a width of 1:

Filename: src/lib.rs

# fn main() {}
#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn larger_can_hold_smaller() {
        let larger = Rectangle { length: 8, width: 7 };
        let smaller = Rectangle { length: 5, width: 1 };

        assert!(larger.can_hold(&smaller));
    }
}

Listing 11-6: A test for can_hold that checks whether a larger rectangle can indeed hold a smaller rectangle

Note that we’ve added a new line inside the tests module: use super::*;. The tests module is a regular module that follows the usual visibility rules we covered in Chapter 7 in the “Privacy Rules” section. Because the tests module is an inner module, we need to bring the code under test in the outer module into the scope of the inner module. We use a glob here so anything we define in the outer module is available to this tests module.

We’ve named our test larger_can_hold_smaller, and we’ve created the two Rectangle instances that we need. Then we called the assert! macro and passed it the result of calling larger.can_hold(&smaller). This expression is supposed to return true, so our test should pass. Let’s find out!

running 1 test
test tests::larger_can_hold_smaller ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

It does pass! Let’s add another test, this time asserting that a smaller rectangle cannot hold a larger rectangle:

Filename: src/lib.rs

# fn main() {}
#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn larger_can_hold_smaller() {
        // --snip--
    }

    #[test]
    fn smaller_cannot_hold_larger() {
        let larger = Rectangle { length: 8, width: 7 };
        let smaller = Rectangle { length: 5, width: 1 };

        assert!(!smaller.can_hold(&larger));
    }
}

Because the correct result of the can_hold function in this case is false, we need to negate that result before we pass it to the assert! macro. As a result, our test will pass if can_hold returns false:

running 2 tests
test tests::smaller_cannot_hold_larger ... ok
test tests::larger_can_hold_smaller ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Two tests that pass! Now let’s see what happens to our test results when we introduce a bug in our code. Let’s change the implementation of the can_hold method by replacing the greater-than sign with a less-than sign when it compares the lengths:

# fn main() {}
# #[derive(Debug)]
# pub struct Rectangle {
#     length: u32,
#     width: u32,
# }
// --snip--

impl Rectangle {
    pub fn can_hold(&self, other: &Rectangle) -> bool {
        self.length < other.length && self.width > other.width
    }
}

Running the tests now produces the following:

running 2 tests
test tests::smaller_cannot_hold_larger ... ok
test tests::larger_can_hold_smaller ... FAILED

failures:

---- tests::larger_can_hold_smaller stdout ----
    thread 'tests::larger_can_hold_smaller' panicked at 'assertion failed:
    larger.can_hold(&smaller)', src/lib.rs:22:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    tests::larger_can_hold_smaller

test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

Our tests caught the bug! Because larger.length is 8 and smaller.length is 5, the comparison of the lengths in can_hold now returns false: 8 is not less than 5.

A common way to test functionality is to compare the result of the code under test to the value you expect the code to return to make sure they’re equal. You could do this using the assert! macro and passing it an expression using the == operator. However, this is such a common test that the standard library provides a pair of macros—assert_eq! and assert_ne!—to perform this test more conveniently. These macros compare two arguments for equality or inequality, respectively. They’ll also print the two values if the assertion fails, which makes it easier to see why the test failed; conversely, the assert! macro only indicates that it got a false value for the == expression, not the values that lead to the false value.

In Listing 11-7, we write a function named add_two that adds 2 to its parameter and returns the result. Then we test this function using the assert_eq! macro.

Filename: src/lib.rs

# fn main() {}
pub fn add_two(a: i32) -> i32 {
    a + 2
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn it_adds_two() {
        assert_eq!(4, add_two(2));
    }
}

Listing 11-7: Testing the function add_two using the assert_eq! macro

Let’s check that it passes!

running 1 test
test tests::it_adds_two ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

The first argument we gave to the assert_eq! macro, 4, is equal to the result of calling add_two(2). The line for this test is test tests::it_adds_two ... ok, and the ok text indicates that our test passed!

Let’s introduce a bug into our code to see what it looks like when a test that uses assert_eq! fails. Change the implementation of the add_two function to instead add 3:

# fn main() {}
pub fn add_two(a: i32) -> i32 {
    a + 3
}

Run the tests again:

running 1 test
test tests::it_adds_two ... FAILED

failures:

---- tests::it_adds_two stdout ----
        thread 'tests::it_adds_two' panicked at 'assertion failed: `(left == right)`
  left: `4`,
 right: `5`', src/lib.rs:11:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    tests::it_adds_two

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

Our test caught the bug! The it_adds_two test failed, displaying the message assertion failed: `(left == right)` and showing that left was 4 and right was 5. This message is useful and helps us start debugging: it means the left argument to assert_eq! was 4 but the right argument, where we had add_two(2), was 5.

Note that in some languages and test frameworks, the parameters to the functions that assert two values are equal are called expected and actual, and the order in which we specify the arguments matters. However, in Rust, they’re called left and right, and the order in which we specify the value we expect and the value that the code under test produces doesn’t matter. We could write the assertion in this test as assert_eq!(add_two(2), 4), which would result in a failure message that displays assertion failed: `(left == right)` and that left was 5 and right was 4.

The assert_ne! macro will pass if the two values we give it are not equal and fail if they’re equal. This macro is most useful for cases when we’re not sure what a value will be, but we know what the value definitely won’t be if our code is functioning as we intend. For example, if we’re testing a function that is guaranteed to change its input in some way, but the way in which the input is changed depends on the day of the week that we run our tests, the best thing to assert might be that the output of the function is not equal to the input.

Under the surface, the assert_eq! and assert_ne! macros use the operators == and !=, respectively. When the assertions fail, these macros print their arguments using debug formatting, which means the values being compared must implement the PartialEq and Debug traits. All the primitive types and most of the standard library types implement these traits. For structs and enums that you define, you’ll need to implement PartialEq to assert that values of those types are equal or not equal. You’ll need to implement Debug to print the values when the assertion fails. Because both traits are derivable traits, as mentioned in Listing 5-12 in Chapter 5, this is usually as straightforward as adding the #[derive(PartialEq, Debug)] annotation to your struct or enum definition. See Appendix C, “Derivable Traits,” for more details about these and other derivable traits.

You can also add a custom message to be printed with the failure message as optional arguments to the assert!, assert_eq!, and assert_ne! macros. Any arguments specified after the one required argument to assert! or the two required arguments to assert_eq! and assert_ne! are passed along to the format! macro (discussed in Chapter 8 in the “Concatenation with the + Operator or the format! Macro” section), so you can pass a format string that contains {} placeholders and values to go in those placeholders. Custom messages are useful to document what an assertion means; when a test fails, you’ll have a better idea of what the problem is with the code.

For example, let’s say we have a function that greets people by name and we want to test that the name we pass into the function appears in the output:

Filename: src/lib.rs

# fn main() {}
pub fn greeting(name: &str) -> String {
    format!("Hello {}!", name)
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn greeting_contains_name() {
        let result = greeting("Carol");
        assert!(result.contains("Carol"));
    }
}

The requirements for this program haven’t been agreed upon yet, and we’re pretty sure the Hello text at the beginning of the greeting will change. We decided we don’t want to have to update the test when the requirements change, so instead of checking for exact equality to the value returned from the greeting function, we’ll just assert that the output contains the text of the input parameter.

Let’s introduce a bug into this code by changing greeting to not include name to see what this test failure looks like:

# fn main() {}
pub fn greeting(name: &str) -> String {
    String::from("Hello!")
}

Running this test produces the following:

running 1 test
test tests::greeting_contains_name ... FAILED

failures:

---- tests::greeting_contains_name stdout ----
        thread 'tests::greeting_contains_name' panicked at 'assertion failed:
result.contains("Carol")', src/lib.rs:12:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    tests::greeting_contains_name

This result just indicates that the assertion failed and which line the assertion is on. A more useful failure message in this case would print the value we got from the greeting function. Let’s change the test function, giving it a custom failure message made from a format string with a placeholder filled in with the actual value we got from the greeting function:

#[test]
fn greeting_contains_name() {
    let result = greeting("Carol");
    assert!(
        result.contains("Carol"),
        "Greeting did not contain name, value was `{}`", result
    );
}

Now when we run the test, we’ll get a more informative error message:

---- tests::greeting_contains_name stdout ----
        thread 'tests::greeting_contains_name' panicked at 'Greeting did not
contain name, value was `Hello!`', src/lib.rs:12:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

We can see the value we actually got in the test output, which would help us debug what happened instead of what we were expecting to happen.

In addition to checking that our code returns the correct values we expect, it’s also important to check that our code handles error conditions as we expect. For example, consider the Guess type that we created in Chapter 9, Listing 9-9. Other code that uses Guess depends on the guarantee that Guess instances will contain only values between 1 and 100. We can write a test that ensures that attempting to create a Guess instance with a value outside that range panics.

We do this by adding another attribute, should_panic, to our test function. This attribute makes a test pass if the code inside the function panics; the test will fail if the code inside the function doesn’t panic.

Listing 11-8 shows a test that checks that the error conditions of Guess::new happen when we expect them to:

Filename: src/lib.rs

# fn main() {}
pub struct Guess {
    value: u32,
}

impl Guess {
    pub fn new(value: u32) -> Guess {
        if value < 1 || value > 100 {
            panic!("Guess value must be between 1 and 100, got {}.", value);
        }

        Guess {
            value
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    #[should_panic]
    fn greater_than_100() {
        Guess::new(200);
    }
}

Listing 11-8: Testing that a condition will cause a panic!

We place the #[should_panic] attribute after the #[test] attribute and before the test function it applies to. Let’s look at the result when this test passes:

running 1 test
test tests::greater_than_100 ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Looks good! Now let’s introduce a bug in our code by removing the condition that the new function will panic if the value is greater than 100:

# fn main() {}
# pub struct Guess {
#     value: u32,
# }
#
// --snip--

impl Guess {
    pub fn new(value: u32) -> Guess {
        if value < 1  {
            panic!("Guess value must be between 1 and 100, got {}.", value);
        }

        Guess {
            value
        }
    }
}

When we run the test in Listing 11-8, it will fail:

running 1 test
test tests::greater_than_100 ... FAILED

failures:

failures:
    tests::greater_than_100

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

We don’t get a very helpful message in this case, but when we look at the test function, we see that it’s annotated with #[should_panic]. The failure we got means that the code in the test function did not cause a panic.

Tests that use should_panic can be imprecise because they only indicate that the code has caused some panic. A should_panic test would pass even if the test panics for a different reason than the one we were expecting to happen. To make should_panic tests more precise, we can add an optional expected parameter to the should_panic attribute. The test harness will make sure that the failure message contains the provided text. For example, consider the modified code for Guess in Listing 11-9 where the new function panics with different messages depending on whether the value is too small or too large:

Filename: src/lib.rs

# fn main() {}
# pub struct Guess {
#     value: u32,
# }
#
// --snip--

impl Guess {
    pub fn new(value: u32) -> Guess {
        if value < 1 {
            panic!("Guess value must be greater than or equal to 1, got {}.",
                   value);
        } else if value > 100 {
            panic!("Guess value must be less than or equal to 100, got {}.",
                   value);
        }

        Guess {
            value
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    #[should_panic(expected = "Guess value must be less than or equal to 100")]
    fn greater_than_100() {
        Guess::new(200);
    }
}

Listing 11-9: Testing that a condition will cause a panic! with a particular panic message

This test will pass because the value we put in the should_panic attribute’s expected parameter is a substring of the message that the Guess::new function panics with. We could have specified the entire panic message that we expect, which in this case would be Guess value must be less than or equal to 100, got 200. What you choose to specify in the expected parameter for should_panic depends on how much of the panic message is unique or dynamic and how precise you want your test to be. In this case, a substring of the panic message is enough to ensure that the code in the test function executes the else if value > 100 case.

To see what happens when a should_panic test with an expected message fails, let’s again introduce a bug into our code by swapping the bodies of the if value < 1 and the else if value > 100 blocks:

if value < 1 {
    panic!("Guess value must be less than or equal to 100, got {}.", value);
} else if value > 100 {
    panic!("Guess value must be greater than or equal to 1, got {}.", value);
}

This time when we run the should_panic test, it will fail:

running 1 test
test tests::greater_than_100 ... FAILED

failures:

---- tests::greater_than_100 stdout ----
        thread 'tests::greater_than_100' panicked at 'Guess value must be
greater than or equal to 1, got 200.', src/lib.rs:11:12
note: Run with `RUST_BACKTRACE=1` for a backtrace.
note: Panic did not include expected string 'Guess value must be less than or
equal to 100'

failures:
    tests::greater_than_100

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

The failure message indicates that this test did indeed panic as we expected, but the panic message did not include the expected string 'Guess value must be less than or equal to 100'. The panic message that we did get in this case was Guess value must be greater than or equal to 1, got 200. Now we can start figuring out where our bug is!

Now that you know several ways to write tests, let’s look at what is happening when we run our tests and explore the different options we can use with cargo test.

Just as cargo run compiles your code and then runs the resulting binary, cargo test compiles your code in test mode and runs the resulting test binary. You can specify command line options to change the default behavior of cargo test. For example, the default behavior of the binary produced by cargo test is to run all the tests in parallel and capture output generated during test runs, preventing the output from being displayed and making it easier to read the output related to the test results.

Some command line options go to cargo test, and some go to the resulting test binary. To separate these two types of arguments, you list the arguments that go to cargo test followed by the separator -- and then the ones that go to the test binary. Running cargo test --help displays the options you can use with cargo test, and running cargo test -- --help displays the options you can use after the separator --.

When you run multiple tests, by default they run in parallel using threads. This means the tests will finish running faster so you can get feedback quicker on whether or not your code is working. Because the tests are running at the same time, make sure your tests don’t depend on each other or on any shared state, including a shared environment, such as the current working directory or environment variables.

For example, say each of your tests runs some code that creates a file on disk named test-output.txt and writes some data to that file. Then each test reads the data in that file and asserts that the file contains a particular value, which is different in each test. Because the tests run at the same time, one test might overwrite the file between when another test writes and reads the file. The second test will then fail, not because the code is incorrect but because the tests have interfered with each other while running in parallel. One solution is to make sure each test writes to a different file; another solution is to run the tests one at a time.

If you don’t want to run the tests in parallel or if you want more fine-grained control over the number of threads used, you can send the --test-threads flag and the number of threads you want to use to the test binary. Take a look at the following example:

$ cargo test -- --test-threads=1

We set the number of test threads to 1, telling the program not to use any parallelism. Running the tests using one thread will take longer than running them in parallel, but the tests won’t interfere with each other if they share state.

By default, if a test passes, Rust’s test library captures anything printed to standard output. For example, if we call println! in a test and the test passes, we won’t see the println! output in the terminal; we’ll see only the line that indicates the test passed. If a test fails, we’ll see whatever was printed to standard output with the rest of the failure message.

As an example, Listing 11-10 has a silly function that prints the value of its parameter and returns 10, as well as a test that passes and a test that fails.

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
fn prints_and_returns_10(a: i32) -> i32 {
    println!("I got the value {}", a);
    10
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn this_test_will_pass() {
        let value = prints_and_returns_10(4);
        assert_eq!(10, value);
    }

    #[test]
    fn this_test_will_fail() {
        let value = prints_and_returns_10(8);
        assert_eq!(5, value);
    }
}
#}

Listing 11-10: Tests for a function that calls println!

When we run these tests with cargo test, we’ll see the following output:

running 2 tests
test tests::this_test_will_pass ... ok
test tests::this_test_will_fail ... FAILED

failures:

---- tests::this_test_will_fail stdout ----
        I got the value 8
thread 'tests::this_test_will_fail' panicked at 'assertion failed: `(left == right)`
  left: `5`,
 right: `10`', src/lib.rs:19:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    tests::this_test_will_fail

test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

Note that nowhere in this output do we see I got the value 4, which is what is printed when the test that passes runs. That output has been captured. The output from the test that failed, I got the value 8, appears in the section of the test summary output, which also shows the cause of the test failure.

If we want to see printed values for passing tests as well, we can disable the output capture behavior by using the --nocapture flag:

$ cargo test -- --nocapture

When we run the tests in Listing 11-10 again with the --nocapture flag, we see the following output:

running 2 tests
I got the value 4
I got the value 8
test tests::this_test_will_pass ... ok
thread 'tests::this_test_will_fail' panicked at 'assertion failed: `(left == right)`
  left: `5`,
 right: `10`', src/lib.rs:19:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.
test tests::this_test_will_fail ... FAILED

failures:

failures:
    tests::this_test_will_fail

test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

Note that the output for the tests and the test results are interleaved; the reason is that the tests are running in parallel, as we talked about in the previous section. Try using the --test-threads=1 option and the --nocapture flag, and see what the output looks like then!

Sometimes, running a full test suite can take a long time. If you’re working on code in a particular area, you might want to run only the tests pertaining to that code. You can choose which tests to run by passing cargo test the name or names of the test(s) you want to run as an argument.

To demonstrate how to run a subset of tests, we’ll create three tests for our add_two function, as shown in Listing 11-11, and choose which ones to run:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub fn add_two(a: i32) -> i32 {
    a + 2
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn add_two_and_two() {
        assert_eq!(4, add_two(2));
    }

    #[test]
    fn add_three_and_two() {
        assert_eq!(5, add_two(3));
    }

    #[test]
    fn one_hundred() {
        assert_eq!(102, add_two(100));
    }
}
#}

Listing 11-11: Three tests with three different names

If we run the tests without passing any arguments, as we saw earlier, all the tests will run in parallel:

running 3 tests
test tests::add_two_and_two ... ok
test tests::add_three_and_two ... ok
test tests::one_hundred ... ok

test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

We can pass the name of any test function to cargo test to run only that test:

$ cargo test one_hundred
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running target/debug/deps/adder-06a75b4a1f2515e9

running 1 test
test tests::one_hundred ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 2 filtered out

Only the test with the name one_hundred ran; the other two tests didn’t match that name. The test output lets us know we had more tests than what this command ran by displaying 2 filtered out at the end of the summary line.

We can’t specify the names of multiple tests in this way; only the first value given to cargo test will be used. But there is a way to run multiple tests.

We can specify part of a test name, and any test whose name matches that value will be run. For example, because two of our tests’ names contain add, we can run those two by running cargo test add:

$ cargo test add
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running target/debug/deps/adder-06a75b4a1f2515e9

running 2 tests
test tests::add_two_and_two ... ok
test tests::add_three_and_two ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 1 filtered out

This command ran all tests with add in the name and filtered out the test named one_hundred. Also note that the module in which tests appear becomes part of the test’s name, so we can run all the tests in a module by filtering on the module’s name.

Sometimes a few specific tests can be very time-consuming to execute, so you might want to exclude them during most runs of cargo test. Rather than listing as arguments all tests you do want to run, you can instead annotate the time-consuming tests using the ignore attribute to exclude them, as shown here:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[test]
fn it_works() {
    assert_eq!(2 + 2, 4);
}

#[test]
#[ignore]
fn expensive_test() {
    // code that takes an hour to run
}
#}

After #[test] we add the #[ignore] line to the test we want to exclude. Now when we run our tests, it_works runs, but expensive_test doesn’t:

$ cargo test
   Compiling adder v0.1.0 (file:///projects/adder)
    Finished dev [unoptimized + debuginfo] target(s) in 0.24 secs
     Running target/debug/deps/adder-ce99bcc2479f4607

running 2 tests
test expensive_test ... ignored
test it_works ... ok

test result: ok. 1 passed; 0 failed; 1 ignored; 0 measured; 0 filtered out

The expensive_test function is listed as ignored. If we want to run only the ignored tests, we can use cargo test -- --ignored:

$ cargo test -- --ignored
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running target/debug/deps/adder-ce99bcc2479f4607

running 1 test
test expensive_test ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 1 filtered out

By controlling which tests run, you can make sure your cargo test results will be fast. When you’re at a point where it makes sense to check the results of the ignored tests and you have time to wait for the results, you can run cargo test -- --ignored instead.

As mentioned at the start of the chapter, testing is a complex discipline, and different people use different terminology and organization. The Rust community thinks about tests in terms of two main categories: unit tests and integration tests. Unit tests are small and more focused, testing one module in isolation at a time, and can test private interfaces. Integration tests are entirely external to your library and use your code in the same way any other external code would, using only the public interface and potentially exercising multiple modules per test.

Writing both kinds of tests is important to ensure that the pieces of your library are doing what you expect them to separately and together.

The purpose of unit tests is to test each unit of code in isolation from the rest of the code to quickly pinpoint where code is and isn’t working as expected. You’ll put unit tests in the src directory in each file with the code that they’re testing. The convention is to create a module named tests in each file to contain the test functions and to annotate the module with cfg(test).

The #[cfg(test)] annotation on the tests module tells Rust to compile and run the test code only when you run cargo test, not when you run cargo build. This saves compile time when you only want to build the library and saves space in the resulting compiled artifact because the tests are not included. You’ll see that because integration tests go in a different directory, they don’t need the #[cfg(test)] annotation. However, because unit tests go in the same files as the code, you’ll use #[cfg(test)] to specify that they shouldn’t be included in the compiled result.

Recall that when we generated the new adder project in the first section of this chapter, Cargo generated this code for us:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
    }
}
#}

This code is the automatically generated test module. The attribute cfg stands for configuration and tells Rust that the following item should only be included given a certain configuration option. In this case, the configuration option is test, which is provided by Rust for compiling and running tests. By using the cfg attribute, Cargo compiles our test code only if we actively run the tests with cargo test. This includes any helper functions that might be within this module, in addition to the functions annotated with #[test].

There’s debate within the testing community about whether or not private functions should be tested directly, and other languages make it difficult or impossible to test private functions. Regardless of which testing ideology you adhere to, Rust’s privacy rules do allow you to test private functions. Consider the code in Listing 11-12 with the private function internal_adder:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub fn add_two(a: i32) -> i32 {
    internal_adder(a, 2)
}

fn internal_adder(a: i32, b: i32) -> i32 {
    a + b
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn internal() {
        assert_eq!(4, internal_adder(2, 2));
    }
}
#}

Listing 11-12: Testing a private function

Note that the internal_adder function is not marked as pub, but because tests are just Rust code and the tests module is just another module, you can import and call internal_adder in a test just fine. If you don’t think private functions should be tested, there’s nothing in Rust that will compel you to do so.

In Rust, integration tests are entirely external to your library. They use your library in the same way any other code would, which means they can only call functions that are part of your library’s public API. Their purpose is to test whether many parts of your library work together correctly. Units of code that work correctly on their own could have problems when integrated, so test coverage of the integrated code is important as well. To create integration tests, you first need a tests directory.

We create a tests directory at the top level of our project directory, next to src. Cargo knows to look for integration test files in this directory. We can then make as many test files as we want to in this directory, and Cargo will compile each of the files as an individual crate.

Let’s create an integration test. With the code in Listing 11-12 still in the src/lib.rs file, make a tests directory, create a new file named tests/integration_test.rs, and enter the code in Listing 11-13:

Filename: tests/integration_test.rs

extern crate adder;

#[test]
fn it_adds_two() {
    assert_eq!(4, adder::add_two(2));
}

Listing 11-13: An integration test of a function in the adder crate

We’ve added extern crate adder at the top of the code, which we didn’t need in the unit tests. The reason is that each test in the tests directory is a separate crate, so we need to import our library into each of them.

We don’t need to annotate any code in tests/integration_test.rs with #[cfg(test)]. Cargo treats the tests directory specially and compiles files in this directory only when we run cargo test. Run cargo test now:

$ cargo test
   Compiling adder v0.1.0 (file:///projects/adder)
    Finished dev [unoptimized + debuginfo] target(s) in 0.31 secs
     Running target/debug/deps/adder-abcabcabc

running 1 test
test tests::internal ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

     Running target/debug/deps/integration_test-ce99bcc2479f4607

running 1 test
test it_adds_two ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

   Doc-tests adder

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

The three sections of output include the unit tests, the integration test, and the doc tests. The first section for the unit tests is the same as we’ve been seeing: one line for each unit test (one named internal that we added in Listing 11-12) and then a summary line for the unit tests.

The integration tests section starts with the line Running target/debug/deps/integration-test-ce99bcc2479f4607 (the hash at the end of your output will be different). Next, there is a line for each test function in that integration test and a summary line for the results of the integration test just before the Doc-tests adder section starts.

Similarly to how adding more unit test functions adds more result lines to the unit tests section, adding more test functions to the integration test file adds more result lines to this integration test file’s section. Each integration test file has its own section, so if we add more files in the tests directory, there will be more integration test sections.

We can still run a particular integration test function by specifying the test function’s name as an argument to cargo test. To run all the tests in a particular integration test file, use the --test argument of cargo test followed by the name of the file:

$ cargo test --test integration_test
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running target/debug/integration_test-952a27e0126bb565

running 1 test
test it_adds_two ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

This command runs only the tests in the tests/integration_test.rs file.

As you add more integration tests, you might want to make more than one file in the tests directory to help organize them; for example, you can group the test functions by the functionality they’re testing. As mentioned earlier, each file in the tests directory is compiled as its own separate crate.

Treating each integration test file as its own crate is useful to create separate scopes that are more like the way end users will be using your crate. However, this means files in the tests directory don’t share the same behavior as files in src do, as you learned in Chapter 7 regarding how to separate code into modules and files.

The different behavior of files in the tests directory is most noticeable when you have a set of helper functions that would be useful in multiple integration test files and you try to follow the steps in the “Moving Modules to Other Files” section of Chapter 7 to extract them into a common module. For example, if we create tests/common.rs and place a function named setup in it, we can add some code to setup that we want to call from multiple test functions in multiple test files:

Filename: tests/common.rs

# #![allow(unused_variables)]
#fn main() {
pub fn setup() {
    // setup code specific to your library's tests would go here
}
#}

When we run the tests again, we’ll see a new section in the test output for the common.rs file, even though this file doesn’t contain any test functions nor did we call the setup function from anywhere:

running 1 test
test tests::internal ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

     Running target/debug/deps/common-b8b07b6f1be2db70

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

     Running target/debug/deps/integration_test-d993c68b431d39df

running 1 test
test it_adds_two ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

   Doc-tests adder

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Having common appear in the test results with running 0 tests displayed for it is not what we wanted. We just wanted to share some code with the other integration test files.

To avoid having common appear in the test output, instead of creating tests/common.rs, we’ll create tests/common/mod.rs. In the “Rules of Module Filesystems” section of Chapter 7, we used the naming convention module_name/mod.rs for files of modules that have submodules. We don’t have submodules for common here, but naming the file this way tells Rust not to treat the common module as an integration test file. When we move the setup function code into tests/common/mod.rs and delete the tests/common.rs file, the section in the test output will no longer appear. Files in subdirectories of the tests directory don’t get compiled as separate crates or have sections in the test output.

After we’ve created tests/common/mod.rs, we can use it from any of the integration test files as a module. Here’s an example of calling the setup function from the it_adds_two test in tests/integration_test.rs:

Filename: tests/integration_test.rs

extern crate adder;

mod common;

#[test]
fn it_adds_two() {
    common::setup();
    assert_eq!(4, adder::add_two(2));
}

Note that the mod common; declaration is the same as the module declarations we demonstrated in Listing 7-4. Then in the test function, we can call the common::setup() function.

If our project is a binary crate that only contains a src/main.rs file and doesn’t have a src/lib.rs file, we can’t create integration tests in the tests directory and use extern crate to import functions defined in the src/main.rs file. Only library crates expose functions that other crates can call and use; binary crates are meant to be run on their own.

This is one of the reasons Rust projects that provide a binary have a straightforward src/main.rs file that calls logic that lives in the src/lib.rs file. Using that structure, integration tests can test the library crate by using extern crate to exercise the important functionality. If the important functionality works, the small amount of code in the src/main.rs file will work as well, and that small amount of code doesn’t need to be tested.

Rust’s testing features provide a way to specify how code should function to ensure it continues to work as you expect, even as you make changes. Unit tests exercise different parts of a library separately and can test private implementation details. Integration tests check that many parts of the library work together correctly, and they use the library’s public API to test the code in the same way external code will use it. Even though Rust’s type system and ownership rules help prevent some kinds of bugs, tests are still important to reduce logic bugs having to do with how your code is expected to behave.

Let’s combine the knowledge you learned in this chapter and in previous chapters to work on a project!

This chapter is a recap of the many skills you’ve learned so far and an exploration of a few more standard library features. We’ll build a command line tool that interacts with file and command line input/output to practice some of the Rust concepts you now have under your belt.

Rust’s speed, safety, single binary output, and cross-platform support make it an ideal language for creating command line tools, so for our project, we’ll make our own version of the classic command line tool grep (globally search a regular expression and print). In the simplest use case, grep searches a specified file for a specified string. To do so, grep takes as its arguments a filename and a string. Then it reads the file, finds lines in that file that contain the string argument, and prints those lines.

Along the way, we’ll show how to make our command line tool use features of the terminal that many command line tools use. We’ll read the value of an environment variable to allow the user to configure the behavior of our tool. We’ll also print to the standard error console stream (stderr) instead of standard output (stdout), so, for example, the user can redirect successful output to a file while still seeing error messages onscreen.

One Rust community member, Andrew Gallant, has already created a fully featured, very fast version of grep, called ripgrep. By comparison, our version of grep will be fairly simple, but this chapter will give you some of the background knowledge you need to understand a real-world project such as ripgrep.

Our grep project will combine a number of concepts you’ve learned so far:

• Organizing code (using what you learned in modules, Chapter 7)
• Using vectors and strings (collections, Chapter 8)
• Handling errors (Chapter 9)
• Using traits and lifetimes where appropriate (Chapter 10)
• Writing tests (Chapter 11)

We’ll also briefly introduce closures, iterators, and trait objects, which Chapters 13 and 17 will cover in detail.

Let’s create a new project with, as always, cargo new. We’ll call our project minigrep to distinguish it from the grep tool that you might already have on your system.

$ cargo new --bin minigrep
     Created binary (application) `minigrep` project
$ cd minigrep

The first task is to make minigrep accept its two command line arguments: the filename and a string to search for. That is, we want to be able to run our program with cargo run, a string to search for, and a path to a file to search in, like so:

$ cargo run searchstring example-filename.txt

Right now, the program generated by cargo new cannot process arguments we give it. Some existing libraries on Crates.io can help with writing a program that accepts command line arguments, but because you’re just learning this concept, let’s implement this capability ourselves.

To enable minigrep to read the values of command line arguments we pass to it, we’ll need a function provided in Rust’s standard library, which is std::env::args. This function returns an iterator of the command line arguments that were given to minigrep. We haven’t discussed iterators yet (we’ll cover them fully in Chapter 13), but for now, you only need to know two details about iterators: iterators produce a series of values, and we can call the collect method on an iterator to turn it into a collection, such as a vector, containing all the elements the iterator produces.

Use the code in Listing 12-1 to allow your minigrep program to read any command line arguments passed to it and then collect the values into a vector:

Filename: src/main.rs

use std::env;

fn main() {
    let args: Vec<String> = env::args().collect();
    println!("{:?}", args);
}

Listing 12-1: Collecting the command line arguments into a vector and printing them

First, we bring the std::env module into scope with a use statement so we can use its args function. Notice that the std::env::args function is nested in two levels of modules. As we discussed in Chapter 7, in cases where the desired function is nested in more than one module, it’s conventional to bring the parent module into scope rather than the function. By doing so, we can easily use other functions from std::env. It’s also less ambiguous than adding use std::env::args and then calling the function with just args, because args might easily be mistaken for a function that’s defined in the current module.

The args Function and Invalid Unicode

Note that std::env::args will panic if any argument contains invalid Unicode. If your program needs to accept arguments containing invalid Unicode, use std::env::args_os instead. That function returns an iterator that produces OsString values instead of String values. We’ve chosen to use std::env::args here for simplicity, because OsString values differ per platform and are more complex to work with than String values.

On the first line of main, we call env::args, and we immediately use collect to turn the iterator into a vector containing all the values produced by the iterator. We can use the collect function to create many kinds of collections, so we explicitly annotate the type of args to specify that we want a vector of strings. Although we very rarely need to annotate types in Rust, collect is one function you do often need to annotate because Rust isn’t able to infer the kind of collection you want.

Finally, we print the vector using the debug formatter, :?. Let’s try running the code first with no arguments and then with two arguments:

$ cargo run
--snip--
["target/debug/minigrep"]

$ cargo run needle haystack
--snip--
["target/debug/minigrep", "needle", "haystack"]

Notice that the first value in the vector is "target/debug/minigrep", which is the name of our binary. This matches the behavior of the arguments list in C, letting programs use the name by which they were invoked in their execution. It’s often convenient to have access to the program name in case you want to print it in messages or change behavior of the program based on what command line alias was used to invoke the program. But for the purposes of this chapter, we’ll ignore it and save only the two arguments we need.

Printing the value of the vector of arguments illustrated that the program is able to access the values specified as command line arguments. Now we need to save the values of the two arguments in variables so we can use the values throughout the rest of the program. We do that in Listing 12-2:

Filename: src/main.rs

use std::env;

fn main() {
    let args: Vec<String> = env::args().collect();

    let query = &args[1];
    let filename = &args[2];

    println!("Searching for {}", query);
    println!("In file {}", filename);
}

Listing 12-2: Creating variables to hold the query argument and filename argument

As we saw when we printed the vector, the program’s name takes up the first value in the vector at args[0], so we’re starting at index 1. The first argument minigrep takes is the string we’re searching for, so we put a reference to the first argument in the variable query. The second argument will be the filename, so we put a reference to the second argument in the variable filename.

We temporarily print the values of these variables to prove that the code is working as we intend. Let’s run this program again with the arguments test and sample.txt:

$ cargo run test sample.txt
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep test sample.txt`
Searching for test
In file sample.txt

Great, the program is working! The values of the arguments we need are being saved into the right variables. Later we’ll add some error handling to deal with certain potential erroneous situations, such as when the user provides no arguments; for now, we’ll ignore that situation and work on adding file-reading capabilities instead.

Now we’ll add functionality to read the file that is specified in the filename command line argument. First, we need a sample file to test it with: the best kind of file to use to make sure minigrep is working is one with a small amount of text over multiple lines with some repeated words. Listing 12-3 has an Emily Dickinson poem that will work well! Create a file called poem.txt at the root level of your project, and enter the poem “I’m Nobody! Who are you?”

Filename: poem.txt

I’m nobody! Who are you?
Are you nobody, too?
Then there’s a pair of us — don’t tell!
They’d banish us, you know.

How dreary to be somebody!
How public, like a frog
To tell your name the livelong day
To an admiring bog!

Listing 12-3: A poem by Emily Dickinson makes a good test case

With the text in place, edit src/main.rs and add code to open the file, as shown in Listing 12-4:

Filename: src/main.rs

use std::env;
use std::fs::File;
use std::io::prelude::*;

fn main() {
#     let args: Vec<String> = env::args().collect();
#
#     let query = &args[1];
#     let filename = &args[2];
#
#     println!("Searching for {}", query);
    // --snip--
    println!("In file {}", filename);

    let mut f = File::open(filename).expect("file not found");

    let mut contents = String::new();
    f.read_to_string(&mut contents)
        .expect("something went wrong reading the file");

    println!("With text:\n{}", contents);
}

Listing 12-4: Reading the contents of the file specified by the second argument

First, we add some more use statements to bring in relevant parts of the standard library: we need std::fs::File to handle files, and std::io::prelude::* contains various useful traits for doing I/O, including file I/O. In the same way that Rust has a general prelude that brings certain types and functions into scope automatically, the std::io module has its own prelude of common types and functions you’ll need when working with I/O. Unlike with the default prelude, we must explicitly add a use statement for the prelude from std::io.

In main, we’ve added three statements: first, we get a mutable handle to the file by calling the File::open function and passing it the value of the filename variable. Second, we create a variable called contents and set it to a mutable, empty String. This will hold the content of the file after we read it in. Third, we call read_to_string on our file handle and pass a mutable reference to contents as an argument.

After those lines, we’ve again added a temporary println! statement that prints the value of contents after the file is read, so we can check that the program is working so far.

Let’s run this code with any string as the first command line argument (because we haven’t implemented the searching part yet) and the poem.txt file as the second argument:

$ cargo run the poem.txt
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep the poem.txt`
Searching for the
In file poem.txt
With text:
I’m nobody! Who are you?
Are you nobody, too?
Then there’s a pair of us — don’t tell!
They’d banish us, you know.

How dreary to be somebody!
How public, like a frog
To tell your name the livelong day
To an admiring bog!

Great! The code read and then printed the contents of the file. But the code has a few flaws. The main function has multiple responsibilities: generally, functions are clearer and easier to maintain if each function is responsible for only one idea. The other problem is that we’re not handling errors as well as we could. The program is still small, so these flaws aren’t a big problem, but as the program grows, it will be harder to fix them cleanly. It’s good practice to begin refactoring early on when developing a program, because it’s much easier to refactor smaller amounts of code. We’ll do that next.

To improve our program, we’ll fix four problems that have to do with the program’s structure and how it’s handling potential errors.

First, our main function now performs two tasks: it parses arguments and opens files. For such a small function, this isn’t a major problem. However, if we continue to grow our program inside main, the number of separate tasks the main function handles will increase. As a function gains responsibilities, it becomes more difficult to reason about, harder to test, and harder to change without breaking one of its parts. It’s best to separate functionality so each function is responsible for one task.

This issue also ties into the second problem: although query and filename are configuration variables to our program, variables like f and contents are used to perform the program’s logic. The longer main becomes, the more variables we’ll need to bring into scope; the more variables we have in scope, the harder it will be to keep track of the purpose of each. It’s best to group the configuration variables into one structure to make their purpose clear.

The third problem is that we’ve used expect to print an error message when opening the file fails, but the error message just prints file not found. Opening a file can fail in a number of ways besides the file being missing: for example, the file might exist, but we might not have permission to open it. Right now, if we’re in that situation, we’d print the file not found error message, which would give the user the wrong information!

Fourth, we use expect repeatedly to handle different errors, and if the user runs our program without specifying enough arguments, they’ll get an index out of bounds error from Rust that doesn’t clearly explain the problem. It would be best if all the error-handling code were in one place so future maintainers had only one place to consult in the code if the error-handling logic needed to change. Having all the error-handling code in one place will also ensure that we’re printing messages that will be meaningful to our end users.

Let’s address these four problems by refactoring our project.

The organizational problem of allocating responsibility for multiple tasks to the main function is common to many binary projects. As a result, the Rust community has developed a process to use as a guideline for splitting the separate concerns of a binary program when main starts getting large. The process has the following steps:

• Split your program into a main.rs and a lib.rs and move your program’s logic to lib.rs.

• As long as your command line parsing logic is small, it can remain in main.rs.

• When the command line parsing logic starts getting complicated, extract it from main.rs and move it to lib.rs.

• The responsibilities that remain in the main function after this process should be limited to the following:

  • Calling the command line parsing logic with the argument values
  • Setting up any other configuration
  • Calling a run function in lib.rs
  • Handling the error if run returns an error

This pattern is about separating concerns: main.rs handles running the program, and lib.rs handles all the logic of the task at hand. Because you can’t test the main function directly, this structure lets you test all of your program’s logic by moving it into functions in lib.rs. The only code that remains in main.rs will be small enough to verify its correctness by reading it. Let’s rework our program by following this process.

We’ll extract the functionality for parsing arguments into a function that main will call to prepare for moving the command line parsing logic to src/lib.rs. Listing 12-5 shows the new start of main that calls a new function parse_config, which we’ll define in src/main.rs for the moment.

Filename: src/main.rs

fn main() {
    let args: Vec<String> = env::args().collect();

    let (query, filename) = parse_config(&args);

    // --snip--
}

fn parse_config(args: &[String]) -> (&str, &str) {
    let query = &args[1];
    let filename = &args[2];

    (query, filename)
}

Listing 12-5: Extracting a parse_config function from main

We’re still collecting the command line arguments into a vector, but instead of assigning the argument value at index 1 to the variable query and the argument value at index 2 to the variable filename within the main function, we pass the whole vector to the parse_config function. The parse_config function then holds the logic that determines which argument goes in which variable and passes the values back to main. We still create the query and filename variables in main, but main no longer has the responsibility of determining how the command line arguments and variables correspond.

This rework may seem like overkill for our small program, but we’re refactoring in small, incremental steps. After making this change, run the program again to verify that the argument parsing still works. It’s good to check your progress often, to help identify the cause of problems when they occur.

We can take another small step to improve the parse_config function further. At the moment, we’re returning a tuple, but then we immediately break that tuple into individual parts again. This is a sign that perhaps we don’t have the right abstraction yet.

Another indicator that shows there’s room for improvement is the config part of parse_config, which implies that the two values we return are related and are both part of one configuration value. We’re not currently conveying this meaning in the structure of the data other than by grouping the two values into a tuple; we could put the two values into one struct and give each of the struct fields a meaningful name. Doing so will make it easier for future maintainers of this code to understand how the different values relate to each other and what their purpose is.

Note: Some people call this anti-pattern of using primitive values when a complex type would be more appropriate primitive obsession.

Listing 12-6 shows the addition of a struct named Config defined to have fields named query and filename. We’ve also changed the parse_config function to return an instance of the Config struct and updated main to use the struct fields rather than having separate variables:

Filename: src/main.rs

# use std::env;
# use std::fs::File;
#
fn main() {
    let args: Vec<String> = env::args().collect();

    let config = parse_config(&args);

    println!("Searching for {}", config.query);
    println!("In file {}", config.filename);

    let mut f = File::open(config.filename).expect("file not found");

    // --snip--
}

struct Config {
    query: String,
    filename: String,
}

fn parse_config(args: &[String]) -> Config {
    let query = args[1].clone();
    let filename = args[2].clone();

    Config { query, filename }
}

Listing 12-6: Refactoring parse_config to return an instance of a Config struct

The signature of parse_config now indicates that it returns a Config value. In the body of parse_config, where we used to return string slices that reference String values in args, we now define Config to contain owned String values. The args variable in main is the owner of the argument values and is only letting the parse_config function borrow them, which means we’d violate Rust’s borrowing rules if Config tried to take ownership of the values in args.

We could manage the String data in a number of different ways, but the easiest, though somewhat inefficient, route is to call the clone method on the values. This will make a full copy of the data for the Config instance to own, which takes more time and memory than storing a reference to the string data. However, cloning the data also makes our code very straightforward because we don’t have to manage the lifetimes of the references; in this circumstance, giving up a little performance to gain simplicity is a worthwhile trade-off.

The Trade-Offs of Using clone

There’s a tendency among many Rustaceans to avoid using clone to fix ownership problems because of its runtime cost. In Chapter 13, you’ll learn how to use more efficient methods in this type of situation. But for now, it’s okay to copy a few strings to continue making progress because you’ll make these copies only once and your filename and query string are very small. It’s better to have a working program that’s a bit inefficient than to try to hyperoptimize code on your first pass. As you become more experienced with Rust, it’ll be easier to start with the most efficient solution, but for now, it’s perfectly acceptable to call clone.

We’ve updated main so it places the instance of Config returned by parse_config into a variable named config, and we updated the code that previously used the separate query and filename variables so it now uses the fields on the Config struct instead.

Now our code more clearly conveys that query and filename are related and that their purpose is to configure how the program will work. Any code that uses these values knows to find them in the config instance in the fields named for their purpose.

So far, we’ve extracted the logic responsible for parsing the command line arguments from main and placed it in the parse_config function. Doing so helped us to see that the query and filename values were related and that relationship should be conveyed in our code. We then added a Config struct to name the related purpose of query and filename and to be able to return the values’ names as struct field names from the parse_config function.

So now that the purpose of the parse_config function is to create a Config instance, we can change parse_config from a plain function to a function named new that is associated with the Config struct. Making this change will make the code more idiomatic. We can create instances of types in the standard library, such as String, by calling String::new. Similarly, by changing parse_config into a new function associated with Config, we’ll be able to create instances of Config by calling Config::new. Listing 12-7 shows the changes we need to make:

Filename: src/main.rs

# use std::env;
#
fn main() {
    let args: Vec<String> = env::args().collect();

    let config = Config::new(&args);

    // --snip--
}

# struct Config {
#     query: String,
#     filename: String,
# }
#
// --snip--

impl Config {
    fn new(args: &[String]) -> Config {
        let query = args[1].clone();
        let filename = args[2].clone();

        Config { query, filename }
    }
}

Listing 12-7: Changing parse_config into Config::new

We’ve updated main where we were calling parse_config to instead call Config::new. We’ve changed the name of parse_config to new and moved it within an impl block, which associates the new function with Config. Try compiling this code again to make sure it works.

Now we’ll work on fixing our error handling. Recall that attempting to access the values in the args vector at index 1 or index 2 will cause the program to panic if the vector contains fewer than three items. Try running the program without any arguments; it will look like this:

$ cargo run
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep`
thread 'main' panicked at 'index out of bounds: the len is 1
but the index is 1', src/main.rs:29:21
note: Run with `RUST_BACKTRACE=1` for a backtrace.

The line index out of bounds: the len is 1 but the index is 1 is an error message intended for programmers. It won’t help our end users understand what happened and what they should do instead. Let’s fix that now.

In Listing 12-8, we add a check in the new function that will verify that the slice is long enough before accessing index 1 and 2. If the slice isn’t long enough, the program panics and displays a better error message than the index out of bounds message.

Filename: src/main.rs

// --snip--
fn new(args: &[String]) -> Config {
    if args.len() < 3 {
        panic!("not enough arguments");
    }
    // --snip--

Listing 12-8: Adding a check for the number of arguments

This code is similar to the Guess::new function we wrote in Listing 9-9, where we called panic! when the value argument was out of the range of valid values. Instead of checking for a range of values here, we’re checking that the length of args is at least 3 and the rest of the function can operate under the assumption that this condition has been met. If args has fewer than three items, this condition will be true, and we call the panic! macro to end the program immediately.

With these extra few lines of code in new, let’s run the program without any arguments again to see what the error looks like now:

$ cargo run
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep`
thread 'main' panicked at 'not enough arguments', src/main.rs:30:12
note: Run with `RUST_BACKTRACE=1` for a backtrace.

This output is better: we now have a reasonable error message. However, we also have extraneous information we don’t want to give to our users. Perhaps using the technique we used in Listing 9-9 isn’t the best to use here: a call to panic! is more appropriate for a programming problem rather than a usage problem, as discussed in Chapter 9. Instead, we can use the other technique you learned about in Chapter 9—returning a Result that indicates either success or an error.

We can instead return a Result value that will contain a Config instance in the successful case and will describe the problem in the error case. When Config::new is communicating to main, we can use the Result type to signal there was a problem. Then we can change main to convert an Err variant into a more practical error for our users without the surrounding text about thread 'main' and RUST_BACKTRACE that a call to panic! causes.

Listing 12-9 shows the changes we need to make to the return value of Config::new and the body of the function needed to return a Result. Note that this won’t compile until we update main as well, which we’ll do in the next listing.

Filename: src/main.rs

impl Config {
    fn new(args: &[String]) -> Result<Config, &'static str> {
        if args.len() < 3 {
            return Err("not enough arguments");
        }

        let query = args[1].clone();
        let filename = args[2].clone();

        Ok(Config { query, filename })
    }
}

Listing 12-9: Returning a Result from Config::new

Our new function now returns a Result with a Config instance in the success case and a &'static str in the error case. Recall from “The Static Lifetime” section in Chapter 10 that &'static str is the type of string literals, which is our error message type for now.

We’ve made two changes in the body of the new function: instead of calling panic! when the user doesn’t pass enough arguments, we now return an Err value, and we’ve wrapped the Config return value in an Ok. These changes make the function conform to its new type signature.

Returning an Err value from Config::new allows the main function to handle the Result value returned from the new function and exit the process more cleanly in the error case.

To handle the error case and print a user-friendly message, we need to update main to handle the Result being returned by Config::new, as shown in Listing 12-10. We’ll also take the responsibility of exiting the command line tool with a nonzero error code from panic! and implement it by hand. A nonzero exit status is a convention to signal to the process that called our program that the program exited with an error state.

Filename: src/main.rs

use std::process;

fn main() {
    let args: Vec<String> = env::args().collect();

    let config = Config::new(&args).unwrap_or_else(|err| {
        println!("Problem parsing arguments: {}", err);
        process::exit(1);
    });

    // --snip--

Listing 12-10: Exiting with an error code if creating a new Config fails

In this listing, we’ve used a method we haven’t covered before: unwrap_or_else, which is defined on Result<T, E> by the standard library. Using unwrap_or_else allows us to define some custom, non-panic! error handling. If the Result is an Ok value, this method’s behavior is similar to unwrap: it returns the inner value Ok is wrapping. However, if the value is an Err value, this method calls the code in the closure, which is an anonymous function we define and pass as an argument to unwrap_or_else. We’ll cover closures in more detail in Chapter 13. For now, you just need to know that unwrap_or_else will pass the inner value of the Err, which in this case is the static string not enough arguments that we added in Listing 12-9, to our closure in the argument err that appears between the vertical pipes. The code in the closure can then use the err value when it runs.

We’ve added a new use line to import process from the standard library. The code in the closure that will be run in the error case is only two lines: we print the err value and then call process::exit. The process::exit function will stop the program immediately and return the number that was passed as the exit status code. This is similar to the panic!-based handling we used in Listing 12-8, but we no longer get all the extra output. Let’s try it:

$ cargo run
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.48 secs
     Running `target/debug/minigrep`
Problem parsing arguments: not enough arguments

Great! This output is much friendlier for our users.

Now that we’ve finished refactoring the configuration parsing, let’s turn to the program’s logic. As we stated in “Separation of Concerns for Binary Projects”, we’ll extract a function named run that will hold all the logic currently in the main function that isn’t involved with setting up configuration or handling errors. When we’re done, main will be concise and easy to verify by inspection, and we’ll be able to write tests for all the other logic.

Listing 12-11 shows the extracted run function. For now, we’re just making the small, incremental improvement of extracting the function. We’re still defining the function in src/main.rs.

Filename: src/main.rs

fn main() {
    // --snip--

    println!("Searching for {}", config.query);
    println!("In file {}", config.filename);

    run(config);
}

fn run(config: Config) {
    let mut f = File::open(config.filename).expect("file not found");

    let mut contents = String::new();
    f.read_to_string(&mut contents)
        .expect("something went wrong reading the file");

    println!("With text:\n{}", contents);
}

// --snip--

Listing 12-11: Extracting a run function containing the rest of the program logic

The run function now contains all the remaining logic from main, starting from reading the file. The run function takes the Config instance as an argument.

With the remaining program logic separated into the run function, we can improve the error handling, as we did with Config::new in Listing 12-9. Instead of allowing the program to panic by calling expect, the run function will return a Result<T, E> when something goes wrong. This will let us further consolidate into main the logic around handling errors in a user-friendly way. Listing 12-12 shows the changes we need to make to the signature and body of run:

Filename: src/main.rs

use std::error::Error;

// --snip--

fn run(config: Config) -> Result<(), Box<Error>> {
    let mut f = File::open(config.filename)?;

    let mut contents = String::new();
    f.read_to_string(&mut contents)?;

    println!("With text:\n{}", contents);

    Ok(())
}

Listing 12-12: Changing the run function to return Result

We’ve made three significant changes here. First, we changed the return type of the run function to Result<(), Box<Error>>. This function previously returned the unit type, (), and we keep that as the value returned in the Ok case.

For the error type, we used the trait object Box<Error> (and we’ve brought std::error::Error into scope with a use statement at the top). We’ll cover trait objects in Chapter 17. For now, just know that Box<Error> means the function will return a type that implements the Error trait, but we don’t have to specify what particular type the return value will be. This gives us flexibility to return error values that may be of different types in different error cases.

Second, we’ve removed the calls to expect in favor of ?, as we talked about in Chapter 9. Rather than panic! on an error, ? will return the error value from the current function for the caller to handle.

Third, the run function now returns an Ok value in the success case. We’ve declared the run function’s success type as () in the signature, which means we need to wrap the unit type value in the Ok value. This Ok(()) syntax might look a bit strange at first, but using () like this is the idiomatic way to indicate that we’re calling run for its side effects only; it doesn’t return a value we need.

When you run this code, it will compile but will display a warning:

warning: unused `std::result::Result` which must be used
  --> src/main.rs:18:5
   |
18 |     run(config);
   |     ^^^^^^^^^^^^
= note: #[warn(unused_must_use)] on by default

Rust tells us that our code ignored the Result value and the Result value might indicate that an error occurred. But we’re not checking to see whether or not there was an error, and the compiler reminds us that we probably meant to have some error handling code here! Let’s rectify that problem now.

We’ll check for errors and handle them using a technique similar to one we used with Config::new in Listing 12-10, but with a slight difference:

Filename: src/main.rs

fn main() {
    // --snip--

    println!("Searching for {}", config.query);
    println!("In file {}", config.filename);

    if let Err(e) = run(config) {
        println!("Application error: {}", e);

        process::exit(1);
    }
}

We use if let rather than unwrap_or_else to check whether run returns an Err value and call process::exit(1) if it does. The run function doesn’t return a value that we want to unwrap in the same way that Config::new returns the Config instance. Because run returns () in the success case, we only care about detecting an error, so we don’t need unwrap_or_else to return the unwrapped value because it would only be ().

The bodies of the if let and the unwrap_or_else functions are the same in both cases: we print the error and exit.

Our minigrep project is looking good so far! Now we’ll split the src/main.rs file and put some code into the src/lib.rs file so we can test it and have a src/main.rs file with fewer responsibilities.

Let’s move all the code that isn’t the main function from src/main.rs to src/lib.rs:

• The run function definition
• The relevant use statements
• The definition of Config
• The Config::new function definition

The contents of src/lib.rs should have the signatures shown in Listing 12-13 (we’ve omitted the bodies of the functions for brevity). Note that this won’t compile until we modify src/main.rs in the listing after this one.

Filename: src/lib.rs

use std::error::Error;
use std::fs::File;
use std::io::prelude::*;

pub struct Config {
    pub query: String,
    pub filename: String,
}

impl Config {
    pub fn new(args: &[String]) -> Result<Config, &'static str> {
        // --snip--
    }
}

pub fn run(config: Config) -> Result<(), Box<Error>> {
    // --snip--
}

Listing 12-13: Moving Config and run into src/lib.rs

We’ve made liberal use of the pub keyword: on Config, on its fields and its new method, and on the run function. We now have a library crate that has a public API that we can test!

Now we need to bring the code we moved to src/lib.rs into the scope of the binary crate in src/main.rs, as shown in Listing 12-14:

Filename: src/main.rs

extern crate minigrep;

use std::env;
use std::process;

use minigrep::Config;

fn main() {
    // --snip--
    if let Err(e) = minigrep::run(config) {
        // --snip--
    }
}

Listing 12-14: Bringing the minigrep crate into the scope of src/main.rs

To bring the library crate into the binary crate, we use extern crate minigrep. Then we add a use minigrep::Config line to bring the Config type into scope, and we prefix the run function with our crate name. Now all the functionality should be connected and should work. Run the program with cargo run and make sure everything works correctly.

Whew! That was a lot of work, but we’ve set ourselves up for success in the future. Now it’s much easier to handle errors, and we’ve made the code more modular. Almost all of our work will be done in src/lib.rs from here on out.

Let’s take advantage of this newfound modularity by doing something that would have been difficult with the old code but is easy with the new code: we’ll write some tests!

Now that we’ve extracted the logic into src/lib.rs and left the argument collecting and error handling in src/main.rs, it’s much easier to write tests for the core functionality of our code. We can call functions directly with various arguments and check return values without having to call our binary from the command line. Feel free to write some tests for the functionality in the Config::new and run functions on your own.

In this section, we’ll add the searching logic to the minigrep program by using the Test-driven development (TDD) process. This software development technique follows these steps:

1. Write a test that fails and run it to make sure it fails for the reason you expect.
2. Write or modify just enough code to make the new test pass.
3. Refactor the code you just added or changed and make sure the tests continue to pass.
4. Repeat from step 1!

This process is just one of many ways to write software, but TDD can help drive code design as well. Writing the test before you write the code that makes the test pass helps to maintain high test coverage throughout the process.

We’ll test drive the implementation of the functionality that will actually do the searching for the query string in the file contents and produce a list of lines that match the query. We’ll add this functionality in a function called search.

Because we don’t need them anymore, let’s remove the println! statements from src/lib.rs and src/main.rs that we used to check the program’s behavior. Then, in src/lib.rs, we’ll add a test module with a test function, as we did in Chapter 11. The test function specifies the behavior we want the search function to have: it will take a query and the text to search for the query in, and it will return only the lines from the text that contain the query. Listing 12-15 shows this test, which won’t compile yet:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
# fn search<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
#      vec![]
# }
#
#[cfg(test)]
mod test {
    use super::*;

    #[test]
    fn one_result() {
        let query = "duct";
        let contents = "\
Rust:
safe, fast, productive.
Pick three.";

        assert_eq!(
            vec!["safe, fast, productive."],
            search(query, contents)
        );
    }
}
#}

Listing 12-15: Creating a failing test for the search function we wish we had

This test searches for the string "duct". The text we’re searching is three lines, only one of which contains "duct". We assert that the value returned from the search function contains only the line we expect.

We aren’t able to run this test and watch it fail because the test doesn’t even compile: the search function doesn’t exist yet! So now we’ll add just enough code to get the test to compile and run by adding a definition of the search function that always returns an empty vector, as shown in Listing 12-16. Then the test should compile and fail because an empty vector doesn’t match a vector containing the line "safe, fast, productive."

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub fn search<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
    vec![]
}
#}

Listing 12-16: Defining just enough of the search function so our test will compile

Notice that we need an explicit lifetime 'a defined in the signature of search and used with the contents argument and the return value. Recall in Chapter 10 that the lifetime parameters specify which argument lifetime is connected to the lifetime of the return value. In this case, we indicate that the returned vector should contain string slices that reference slices of the argument contents (rather than the argument query).

In other words, we tell Rust that the data returned by the search function will live as long as the data passed into the search function in the contents argument. This is important! The data referenced by a slice needs to be valid for the reference to be valid; if the compiler assumes we’re making string slices of query rather than contents, it will do its safety checking incorrectly.

If we forget the lifetime annotations and try to compile this function, we’ll get this error:

error[E0106]: missing lifetime specifier
 --> src/lib.rs:5:51
  |
5 | pub fn search(query: &str, contents: &str) -> Vec<&str> {
  |                                                   ^ expected lifetime
parameter
  |
  = help: this function's return type contains a borrowed value, but the
  signature does not say whether it is borrowed from `query` or `contents`

Rust can’t possibly know which of the two arguments we need, so we need to tell it. Because contents is the argument that contains all of our text and we want to return the parts of that text that match, we know contents is the argument that should be connected to the return value using the lifetime syntax.

Other programming languages don’t require you to connect arguments to return values in the signature. So although this might seem strange, it will get easier over time. You might want to compare this example with the “Validating References with Lifetimes” section in Chapter 10.

Now let’s run the test:

$ cargo test
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
--warnings--
    Finished dev [unoptimized + debuginfo] target(s) in 0.43 secs
     Running target/debug/deps/minigrep-abcabcabc

running 1 test
test test::one_result ... FAILED

failures:

---- test::one_result stdout ----
        thread 'test::one_result' panicked at 'assertion failed: `(left ==
right)`
left: `["safe, fast, productive."]`,
right: `[]`)', src/lib.rs:48:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.


failures:
    test::one_result

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

error: test failed, to rerun pass '--lib'

Great, the test fails, exactly as we expected. Let’s get the test to pass!

Currently, our test is failing because we always return an empty vector. To fix that and implement search, our program needs to follow these steps:

• Iterate through each line of the contents.
• Check whether the line contains our query string.
• If it does, add it to the list of values we’re returning.
• If it doesn’t, do nothing.
• Return the list of results that match.

Let’s work through each step, starting with iterating through lines.

Rust has a helpful method to handle line-by-line iteration of strings, conveniently named lines, that works as shown in Listing 12-17. Note this won’t compile yet:

Filename: src/lib.rs

pub fn search<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
    for line in contents.lines() {
        // do something with line
    }
}

Listing 12-17: Iterating through each line in contents

The lines method returns an iterator. We’ll talk about iterators in depth in Chapter 13, but recall that you saw this way of using an iterator in Listing 3-5, where we used a for loop with an iterator to run some code on each item in a collection.

Next, we’ll check whether the current line contains our query string. Fortunately, strings have a helpful method named contains that does this for us! Add a call to the contains method in the search function, as shown in Listing 12-18. Note this still won’t compile yet:

Filename: src/lib.rs

pub fn search<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
    for line in contents.lines() {
        if line.contains(query) {
            // do something with line
        }
    }
}

Listing 12-18: Adding functionality to see whether the line contains the string in query

We also need a way to store the lines that contain our query string. For that, we can make a mutable vector before the for loop and call the push method to store a line in the vector. After the for loop, we return the vector, as shown in Listing 12-19:

Filename: src/lib.rs

pub fn search<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
    let mut results = Vec::new();

    for line in contents.lines() {
        if line.contains(query) {
            results.push(line);
        }
    }

    results
}

Listing 12-19: Storing the lines that match so we can return them

Now the search function should return only the lines that contain query, and our test should pass. Let’s run the test:

$ cargo test
--snip--
running 1 test
test test::one_result ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Our test passed, so we know it works!

At this point, we could consider opportunities for refactoring the implementation of the search function while keeping the tests passing to maintain the same functionality. The code in the search function isn’t too bad, but it doesn’t take advantage of some useful features of iterators. We’ll return to this example in Chapter 13, where we’ll explore iterators in detail, and look at how to improve it.

Now that the search function is working and tested, we need to call search from our run function. We need to pass the config.query value and the contents that run reads from the file to the search function. Then run will print each line returned from search:

Filename: src/lib.rs

pub fn run(config: Config) -> Result<(), Box<Error>> {
    let mut f = File::open(config.filename)?;

    let mut contents = String::new();
    f.read_to_string(&mut contents)?;

    for line in search(&config.query, &contents) {
        println!("{}", line);
    }

    Ok(())
}

We’re still using a for loop to return each line from search and print it.

Now the entire program should work! Let’s try it out, first with a word that should return exactly one line from the Emily Dickinson poem, “frog”:

$ cargo run frog poem.txt
   Compiling minigrep v0.1.0 (file:///projects/minigrep)
    Finished dev [unoptimized + debuginfo] target(s) in 0.38 secs
     Running `target/debug/minigrep frog poem.txt`
How public, like a frog

Cool! Now let’s try a word that will match multiple lines, like “body”:

$ cargo run body poem.txt
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep body poem.txt`
I’m nobody! Who are you?
Are you nobody, too?
How dreary to be somebody!

And finally, let’s make sure that we don’t get any lines when we search for a word that isn’t anywhere in the poem, such as “monomorphization”:

$ cargo run monomorphization poem.txt
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/minigrep monomorphization poem.txt`

Excellent! We’ve built our own mini version of a classic tool and learned a lot about how to structure applications. We’ve also learned a bit about file input and output, lifetimes, testing, and command line parsing.

To round out this project, we’ll briefly demonstrate how to work with environment variables and how to print to standard error, both of which are useful when you’re writing command line programs.

We’ll improve minigrep by adding an extra feature: an option for case-insensitive searching that the user can turn on via an environment variable. We could make this feature a command line option and require that users enter it each time they want it to apply, but instead we’ll use an environment variable. Doing so allows our users to set the environment variable once and have all their searches be case insensitive in that terminal session.

We want to add a new search_case_insensitive function that we’ll call when the environment variable is on. We’ll continue to follow the TDD process, so the first step is again to write a failing test. We’ll add a new test for the new search_case_insensitive function and rename our old test from one_result to case_sensitive to clarify the differences between the two tests, as shown in Listing 12-20:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
#[cfg(test)]
mod test {
    use super::*;

    #[test]
    fn case_sensitive() {
        let query = "duct";
        let contents = "\
Rust:
safe, fast, productive.
Pick three.
Duct tape.";

        assert_eq!(
            vec!["safe, fast, productive."],
            search(query, contents)
        );
    }

    #[test]
    fn case_insensitive() {
        let query = "rUsT";
        let contents = "\
Rust:
safe, fast, productive.
Pick three.
Trust me.";

        assert_eq!(
            vec!["Rust:", "Trust me."],
            search_case_insensitive(query, contents)
        );
    }
}
#}

Listing 12-20: Adding a new failing test for the case-insensitive function we’re about to add

Note that we’ve edited the old test’s contents too. We’ve added a new line with the text "Duct tape." using a capital D that shouldn’t match the query “duct” when we’re searching in a case-sensitive manner. Changing the old test in this way helps ensure that we don’t accidentally break the case-sensitive search functionality that we’ve already implemented. This test should pass now and should continue to pass as we work on the case-insensitive search.

The new test for the case-insensitive search uses "rUsT" as its query. In the search_case_insensitive function we’re about to add, the query "rUsT" should match the line containing "Rust:" with a capital R and match the line "Trust me." even though both have different casing than the query. This is our failing test, and it will fail to compile because we haven’t yet defined the search_case_insensitive function. Feel free to add a skeleton implementation that always returns an empty vector, similar to the way we did for the search function in Listing 12-16 to see the test compile and fail.

The search_case_insensitive function, shown in Listing 12-21, will be almost the same as the search function. The only difference is that we’ll lowercase the query and each line so whatever the case of the input arguments, they’ll be the same case when we check whether the line contains the query.

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
fn search_case_insensitive<'a>(query: &str, contents: &'a str) -> Vec<&'a str> {
    let query = query.to_lowercase();
    let mut results = Vec::new();

    for line in contents.lines() {
        if line.to_lowercase().contains(&query) {
            results.push(line);
        }
    }

    results
}
#}

Listing 12-21: Defining the search_case_insensitive function to lowercase the query and the line before comparing them

First, we lowercase the query string and store it in a shadowed variable with the same name. Calling to_lowercase on the query is necessary so no matter whether the user’s query is "rust", "RUST", "Rust:", or "rUsT", we’ll treat the query as if it were "rust" and be insensitive to the case.

Note that query is now a String rather than a string slice, because calling to_lowercase creates new data rather than referencing existing data. Say the query is "rUsT", as an example: that string slice doesn’t contain a lowercase u or t for us to use, so we have to allocate a new String containing "rust". When we pass query as an argument to the contains method now, we need to add an ampersand because the signature of contains is defined to take a string slice.

Next, we add a call to to_lowercase on each line before we check whether it contains query to lowercase all characters. Now that we’ve converted line and query to lowercase, we’ll find matches no matter what the case of the query is.

Let’s see if this implementation passes the tests:

running 2 tests
test test::case_insensitive ... ok
test test::case_sensitive ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Great! They passed. Now, let’s call the new search_case_insensitive function from the run function. First, we’ll add a configuration option to the Config struct to switch between case-sensitive and case-insensitive search. Adding this field will cause compiler errors since we aren’t initializing this field anywhere yet:

Filename: src/lib.rs

# #![allow(unused_variables)]
#fn main() {
pub struct Config {
    pub query: String,
    pub filename: String,
    pub case_sensitive: bool,
}
#}

Note that we added the case_sensitive field that holds a Boolean. Next, we need the run function to check the case_sensitive field’s value and use that to decide whether to call the search function or the search_case_insensitive function, as shown in Listing 12-22. Note this still won’t compile yet: